[{"data":1,"prerenderedAt":1690},["ShallowReactive",2],{"article-ai-workflow-automation":3,"related-articles-ai-workflow-automation":337},{"id":4,"title":5,"author":6,"body":7,"category":309,"description":310,"extension":311,"featured":312,"heroImage":313,"meta":314,"navigation":312,"path":315,"publishedAt":316,"readingTime":317,"relatedAgents":318,"relatedArticles":322,"relatedWorkflows":326,"searchIntent":329,"seo":330,"stem":331,"topics":332,"updatedAt":316,"visual":21,"__hash__":336},"articles\u002Farticles\u002Fai-workflow-automation.md","AI workflow automation: automate the rules, not the judgment","StackOS team",{"type":8,"value":9,"toc":299},"minimark",[10,14,17,23,28,31,34,37,40,43,46,50,72,75,78,100,103,107,110,113,116,119,122,130,134,137,140,238,241,244,248,251,254,257,265,269,272,290,293,296],[11,12,13],"p",{},"AI workflow automation combines two different kinds of control. Code should enforce the rules that must hold every time: state, dependencies, permissions, schemas, receipts, and stopping conditions. An agent should decide what cannot be known until the work is underway: which evidence matters, whether the plan should adapt, and which feedback belongs in the result.",[11,15,16],{},"Treating the whole workflow as either a fixed automation or an autonomous agent misses the useful middle. The practical design question is not, “How much AI can we add?” It is, “Which decisions should still require judgment?”",[18,19],"article-concept-visual",{"caption":20,"mode":21,"title":22},"The runtime enforces invariants. Agents make evidence-dependent decisions. People enter for missing intent or consequential choices.","connections","One workflow, three kinds of control",[24,25,27],"h2",{"id":26},"a-failure-that-a-better-prompt-could-not-fix","A failure that a better prompt could not fix",[11,29,30],{},"We ran into a useful example while refining our own content workflow.",[11,32,33],{},"The workflow definition had changed, but part of the running system still held an older generation of the plugin and resource contracts. The agent could see the current content workflow while receiving an older schema for the record it needed to write. One part of the system described the right job; another enforced the wrong shape.",[11,35,36],{},"There was a second gap. Each workflow step already declared an output contract, but a step could still be recorded as successful without proving that its result matched that contract.",[11,38,39],{},"Neither problem belonged in the prompt. Telling the agent to “use the latest schema” would not make two runtime generations consistent. Telling it to “return every required field” would not make success verifiable.",[11,41,42],{},"We moved both responsibilities into the mechanical layer. Editable plugin manifests now carry a source-generation fingerprint, and the catalog resynchronizes when that generation changes. Successful step results are validated against their frozen JSON Schema before any lifecycle transition is persisted. If a required field is missing, the step remains running and the error points to the exact field that needs repair.",[11,44,45],{},"We then exercised the failure path live: an incomplete result was rejected without advancing the step, and the corrected result completed it. This is bounded first-party evidence from one implementation, not proof that the same architecture fits every system. The lesson is narrower: if correctness depends on an invariant, the system should enforce it without asking the model to remember.",[24,47,49],{"id":48},"put-stable-rules-in-the-mechanical-layer","Put stable rules in the mechanical layer",[11,51,52,59,60,65,66,71],{},[53,54,58],"a",{"href":55,"rel":56},"https:\u002F\u002Fwww.anthropic.com\u002Fengineering\u002Fbuilding-effective-agents",[57],"nofollow","Anthropic distinguishes workflows from agents"," by who controls the path: workflows use predefined code paths, while agents dynamically direct their process and tool use. Microsoft’s ",[53,61,64],{"href":62,"rel":63},"https:\u002F\u002Flearn.microsoft.com\u002Fen-us\u002Fazure\u002Farchitecture\u002Fai-ml\u002Fguide\u002Fai-agent-design-patterns",[57],"orchestration pattern guidance"," and Google Cloud’s ",[53,67,70],{"href":68,"rel":69},"https:\u002F\u002Fdocs.cloud.google.com\u002Farchitecture\u002Fchoose-design-pattern-agentic-ai-system",[57],"agentic design-pattern guide"," make a similar distinction between predictable sequences and dynamic orchestration.",[11,73,74],{},"That distinction is useful inside one workflow, not only when choosing an architecture for the whole system.",[11,76,77],{},"The mechanical layer should own conditions whose answer does not improve with another model call:",[79,80,81,85,88,91,94,97],"ul",{},[82,83,84],"li",{},"A dependent step cannot start before its prerequisite finishes.",[82,86,87],{},"A research step can read sources but cannot publish.",[82,89,90],{},"A successful result must contain the fields and types its consumer expects.",[82,92,93],{},"A completed external action needs a receipt the workflow can inspect before retrying.",[82,95,96],{},"A packet-only request must stop before publication.",[82,98,99],{},"The current state and evidence refs must survive the conversation that produced them.",[11,101,102],{},"These controls reduce ambiguity without reducing useful autonomy. The agent no longer spends judgment on whether a required field is optional this time or whether “packet only” might permit a deployment. It can use that judgment on the work itself.",[24,104,106],{"id":105},"keep-evidence-dependent-choices-with-the-agent","Keep evidence-dependent choices with the agent",[11,108,109],{},"Some decisions look repetitive but do not have a stable answer.",[11,111,112],{},"In this article’s workflow, the interview step was always present, but the interview was not mandatory. The agent had to inspect the current voice guide, prior pieces, recorded operator statements, and the new topic. Those sources already captured the relevant judgment, so the interview was skipped with a reason and a list of perspectives the article could not claim.",[11,114,115],{},"Hard-coding “always interview” would add ceremony. Hard-coding “never interview” would remove a source when first-hand experience was actually missing. The stable rule is that the decision must be made and explained. The answer remains contextual.",[11,117,118],{},"The same boundary applies to research and review. A schema can require a source ledger; it cannot decide which source resolves a contradiction. A workflow can require independent claim and voice review; it should not automatically turn every reviewer preference into another delivery cycle.",[11,120,121],{},"The orchestrator owns that gate. It can accept a supported blocker, send a specific defect back for repair, retain a preference as advice, and reject an unsupported or out-of-scope finding. Review produces evidence. It does not acquire authority over the original goal merely because it happened later.",[11,123,124,125,129],{},"This is where an ",[53,126,128],{"href":127},"\u002Flibrary\u002Farticles\u002Fhow-to-build-ai-agent-workflow","AI agent workflow"," differs from a long automation script. The contract defines the operating boundary. Reasoning handles the parts whose answer depends on meaning, evidence, or changed conditions.",[24,131,133],{"id":132},"one-workflow-can-mix-all-three-control-modes","One workflow can mix all three control modes",[11,135,136],{},"Our content workflow follows a visible sequence: decide on interview scope, collect evidence, choose an angle, draft, review claims and voice, check disclosure risk, render the selected channel packet, preserve the final record, and stop or publish according to the request.",[11,138,139],{},"The sequence and handoffs are deterministic. The work inside them is not.",[141,142,143,159],"table",{},[144,145,146],"thead",{},[147,148,149,153,156],"tr",{},[150,151,152],"th",{},"Responsibility",[150,154,155],{},"Best owner",[150,157,158],{},"Why",[160,161,162,174,185,196,206,217,227],"tbody",{},[147,163,164,168,171],{},[165,166,167],"td",{},"Require a source ledger and claim map",[165,169,170],{},"Runtime contract",[165,172,173],{},"The requirement is stable and machine-checkable.",[147,175,176,179,182],{},[165,177,178],{},"Decide whether existing evidence is sufficient",[165,180,181],{},"Agent",[165,183,184],{},"The answer depends on the topic, source quality, and claims being considered.",[147,186,187,190,193],{},[165,188,189],{},"Prevent drafting before research completes",[165,191,192],{},"Workflow state",[165,194,195],{},"The dependency should hold on every run.",[147,197,198,201,203],{},[165,199,200],{},"Choose the article angle",[165,202,181],{},[165,204,205],{},"Reader value and evidence fit require interpretation.",[147,207,208,211,214],{},[165,209,210],{},"Decide whether a review finding changes delivery",[165,212,213],{},"Orchestrator",[165,215,216],{},"The finding must be tested against the accepted goal and evidence.",[147,218,219,222,224],{},[165,220,221],{},"Prevent a packet-only run from publishing",[165,223,170],{},[165,225,226],{},"Execution intent is an explicit boundary, not a suggestion.",[147,228,229,232,235],{},[165,230,231],{},"Clarify a missing destination or sensitive disclosure choice",[165,233,234],{},"Person",[165,236,237],{},"Guessing would materially change the requested action or public boundary.",[11,239,240],{},"The person is not a rubber stamp between every row. Human participation belongs where the system lacks legitimate authority or information: the goal is ambiguous, evidence conflicts on a consequential point, disclosure ownership is unclear, or an external action needs a choice that was never supplied.",[11,242,243],{},"That is different from inserting approval because AI is involved. A mandatory checkpoint can be appropriate for a risky action. It is not a substitute for a clear workflow.",[24,245,247],{"id":246},"use-failure-location-to-improve-the-boundary","Use failure location to improve the boundary",[11,249,250],{},"When an AI workflow fails, ask which layer was forced to compensate.",[11,252,253],{},"If the agent searched for a source the workflow already knew, context selection failed. If it guessed which account or destination to use, authority was unresolved. If it returned a malformed packet and the system accepted it, validation failed. If a reviewer’s optional rewrite expanded the task, feedback adjudication failed. If the system stopped on missing operator intent and asked one focused question, the boundary may have worked exactly as intended.",[11,255,256],{},"This makes ordinary friction useful evidence. A workflow does not need to eliminate every search, retry, or clarification. It needs to keep recovery local and prevent that friction from turning into silent drift.",[11,258,259,260,264],{},"The ",[53,261,263],{"href":262},"\u002Flibrary\u002Farticles\u002Fai-agent-experience","agent experience"," article develops that point at the claimed-step level. The automation boundary is the system-level version: decide which uncertainty the agent should resolve and which uncertainty the system should remove before the step begins.",[24,266,268],{"id":267},"a-compact-automation-boundary-test","A compact automation-boundary test",[11,270,271],{},"For each responsibility in a workflow, ask:",[273,274,275,278,281,284,287],"ol",{},[82,276,277],{},"Does the rule need to hold on every valid run?",[82,279,280],{},"Can the result be checked without interpreting meaning?",[82,282,283],{},"Does the answer change with evidence or intermediate results?",[82,285,286],{},"Would a wrong guess change scope, expose data, spend money, or create an external side effect?",[82,288,289],{},"Can a failed attempt be repaired locally without replaying completed work?",[11,291,292],{},"Stable and machine-checkable responsibilities belong in code, schemas, state transitions, or scoped tool grants. Evidence-dependent responsibilities belong with a reasoning agent. Missing authority or a genuinely consequential choice belongs with the person who owns it.",[11,294,295],{},"The boundary will not be perfect on the first run. Ours was not. The useful signal was that a live failure identified two invariants the runtime had left to convention. We fixed those invariants mechanically and kept the editorial decisions with the agents.",[11,297,298],{},"That is the point of AI workflow automation: not to automate judgment away, but to stop wasting it on rules the system can already know.",{"title":300,"searchDepth":301,"depth":301,"links":302},"",2,[303,304,305,306,307,308],{"id":26,"depth":301,"text":27},{"id":48,"depth":301,"text":49},{"id":105,"depth":301,"text":106},{"id":132,"depth":301,"text":133},{"id":246,"depth":301,"text":247},{"id":267,"depth":301,"text":268},"AI operations","A practical way to decide what AI workflows should enforce in code, what agents should decide at runtime, and when a person genuinely needs to step in.","md",true,null,{},"\u002Farticles\u002Fai-workflow-automation","2026-07-12","8 min read",[319,320,321],"branding-channel-strategist","branding-narrative-writer","branding-claim-auditor",[323,324,325],"how-to-build-ai-agent-workflow","ai-agent-experience","what-is-an-agentic-workflow",[327,328],"branding-content-production","engineering-tracked-delivery","Learn how to automate AI workflows without hard-coding the judgment they need",{"title":5,"description":310},"articles\u002Fai-workflow-automation",[333,334,335],"AI workflow automation","agent orchestration","workflow architecture","ivj4k_-yJoUYW08VqRr6mbUEC2lYzjoWMWbJCgOnxvo",[338,538,808,1108,1248,1378,1512],{"id":4,"title":5,"author":6,"body":339,"category":309,"description":310,"extension":311,"featured":312,"heroImage":313,"meta":532,"navigation":312,"path":315,"publishedAt":316,"readingTime":317,"relatedAgents":533,"relatedArticles":534,"relatedWorkflows":535,"searchIntent":329,"seo":536,"stem":331,"topics":537,"updatedAt":316,"visual":21,"__hash__":336},{"type":8,"value":340,"toc":524},[341,343,345,347,349,351,353,355,357,359,361,363,374,376,378,392,394,396,398,400,402,404,406,410,412,414,416,486,488,490,492,494,496,498,502,504,506,518,520,522],[11,342,13],{},[11,344,16],{},[18,346],{"caption":20,"mode":21,"title":22},[24,348,27],{"id":26},[11,350,30],{},[11,352,33],{},[11,354,36],{},[11,356,39],{},[11,358,42],{},[11,360,45],{},[24,362,49],{"id":48},[11,364,365,59,368,65,371,71],{},[53,366,58],{"href":55,"rel":367},[57],[53,369,64],{"href":62,"rel":370},[57],[53,372,70],{"href":68,"rel":373},[57],[11,375,74],{},[11,377,77],{},[79,379,380,382,384,386,388,390],{},[82,381,84],{},[82,383,87],{},[82,385,90],{},[82,387,93],{},[82,389,96],{},[82,391,99],{},[11,393,102],{},[24,395,106],{"id":105},[11,397,109],{},[11,399,112],{},[11,401,115],{},[11,403,118],{},[11,405,121],{},[11,407,124,408,129],{},[53,409,128],{"href":127},[24,411,133],{"id":132},[11,413,136],{},[11,415,139],{},[141,417,418,428],{},[144,419,420],{},[147,421,422,424,426],{},[150,423,152],{},[150,425,155],{},[150,427,158],{},[160,429,430,438,446,454,462,470,478],{},[147,431,432,434,436],{},[165,433,167],{},[165,435,170],{},[165,437,173],{},[147,439,440,442,444],{},[165,441,178],{},[165,443,181],{},[165,445,184],{},[147,447,448,450,452],{},[165,449,189],{},[165,451,192],{},[165,453,195],{},[147,455,456,458,460],{},[165,457,200],{},[165,459,181],{},[165,461,205],{},[147,463,464,466,468],{},[165,465,210],{},[165,467,213],{},[165,469,216],{},[147,471,472,474,476],{},[165,473,221],{},[165,475,170],{},[165,477,226],{},[147,479,480,482,484],{},[165,481,231],{},[165,483,234],{},[165,485,237],{},[11,487,240],{},[11,489,243],{},[24,491,247],{"id":246},[11,493,250],{},[11,495,253],{},[11,497,256],{},[11,499,259,500,264],{},[53,501,263],{"href":262},[24,503,268],{"id":267},[11,505,271],{},[273,507,508,510,512,514,516],{},[82,509,277],{},[82,511,280],{},[82,513,283],{},[82,515,286],{},[82,517,289],{},[11,519,292],{},[11,521,295],{},[11,523,298],{},{"title":300,"searchDepth":301,"depth":301,"links":525},[526,527,528,529,530,531],{"id":26,"depth":301,"text":27},{"id":48,"depth":301,"text":49},{"id":105,"depth":301,"text":106},{"id":132,"depth":301,"text":133},{"id":246,"depth":301,"text":247},{"id":267,"depth":301,"text":268},{},[319,320,321],[323,324,325],[327,328],{"title":5,"description":310},[333,334,335],{"id":539,"title":540,"author":6,"body":541,"category":309,"description":791,"extension":311,"featured":312,"heroImage":313,"meta":792,"navigation":312,"path":793,"publishedAt":794,"readingTime":795,"relatedAgents":796,"relatedArticles":798,"relatedWorkflows":800,"searchIntent":801,"seo":802,"stem":803,"topics":804,"updatedAt":794,"visual":552,"__hash__":807},"articles\u002Farticles\u002Fai-agent-experience.md","Agent experience: the missing layer in AI agent orchestration",{"type":8,"value":542,"toc":783},[543,546,549,554,557,560,568,572,575,578,581,589,592,596,599,602,665,668,671,674,677,680,684,687,690,693,696,699,703,706,709,712,715,719,722,725,728,731,734,738,741,744,767,770,777,780],[11,544,545],{},"An orchestration plan can be logically correct and still fail at the moment an agent receives its next step. The workflow says “draft the article,” but the agent must discover the brief, infer the audience, locate the allowed tools, reconstruct prior decisions, decide what completion means, and determine where the result belongs.",[11,547,548],{},"That gap lives in the agent’s operating environment at the point of action.",[18,550],{"caption":551,"mode":552,"title":553},"Orchestration is experienced one claimed step at a time: context, authority, tools, outputs, recovery, and a stopping rule.","roles","The workflow an agent actually receives",[11,555,556],{},"Agent experience is the per-step operating layer that makes an orchestration plan executable for a fresh agent. Operationally, it is the burden a system places on that agent to search, guess, rediscover, and recover before it can perform valid work.",[11,558,559],{},"This definition keeps the idea concrete. Search burden is time spent locating relevant state or instructions. Guessing burden appears when inputs, authority, or completion criteria are ambiguous. Rediscovery burden comes from reconstructing decisions the system already knows. Recovery burden is the work required to understand and repair a failed attempt.",[11,561,562,563,567],{},"These burdens are separate from the distinctions among an ",[53,564,566],{"href":565},"\u002Flibrary\u002Farticles\u002Fai-agent-vs-workflow-vs-orchestrator","AI agent, workflow, and orchestrator",". Those components may all be present while the claimed step remains difficult to execute.",[24,569,571],{"id":570},"an-assignment-is-not-yet-an-executable-step","An assignment is not yet an executable step",[11,573,574],{},"“Review the implementation” is an assignment. It identifies an activity but leaves most operating questions unanswered.",[11,576,577],{},"Which implementation? Review against which requirements? May the reviewer run tests, inspect external systems, or modify files? What evidence should the review produce? Does “ready” mean no defects, no blocking defects, or acceptance of documented risk? If the review finds a problem, who receives it and in what form?",[11,579,580],{},"A capable agent can often fill these gaps. That is precisely the problem: successful execution now depends on inference that the orchestration system could have resolved before dispatch.",[11,582,583,584,588],{},"An executable step should let an agent move from claim to first valid action without rebuilding the workflow in its own context window. This is especially important in an ",[53,585,587],{"href":586},"\u002Flibrary\u002Farticles\u002Fwhat-is-an-agentic-workflow","agentic workflow",", where later actions depend on runtime findings. Dynamic behavior increases the need for explicit local operating conditions; it does not remove it.",[11,590,591],{},"The claimed step is therefore the useful unit for examining agent experience.",[24,593,595],{"id":594},"the-claimed-step-packet","The claimed-step packet",[11,597,598],{},"When an agent claims a step, it should receive a bounded packet containing what that step needs now. This prepared execution surface selects from project state instead of passing all of it through.",[11,600,601],{},"A useful packet includes:",[79,603,604,611,617,623,629,635,641,647,653,659],{},[82,605,606,610],{},[607,608,609],"strong",{},"Purpose:"," why the work exists and what downstream decision or action it supports.",[82,612,613,616],{},[607,614,615],{},"Instructions and policies:"," the relevant rules, already selected for this step.",[82,618,619,622],{},[607,620,621],{},"Completion criteria:"," observable conditions that distinguish finished work from plausible-looking progress.",[82,624,625,628],{},[607,626,627],{},"Exact tools:"," callable operations, expected use, and important limitations.",[82,630,631,634],{},[607,632,633],{},"Resolved inputs:"," concrete artifact references, resource identifiers, prior outputs, and configuration values rather than instructions to “find the latest.”",[82,636,637,640],{},[607,638,639],{},"Bounded context:"," enough history to understand the work without replaying the whole run.",[82,642,643,646],{},[607,644,645],{},"Output contracts:"," required fields, artifact formats, evidence, and status semantics.",[82,648,649,652],{},[607,650,651],{},"Direct handoff:"," where the result goes and what the next actor needs from it.",[82,654,655,658],{},[607,656,657],{},"Targeted recovery:"," likely failure modes and the next safe action for each one.",[82,660,661,664],{},[607,662,663],{},"A stopping rule:"," when to return control instead of expanding the task.",[11,666,667],{},"These fields turn hidden orchestration knowledge into local operating knowledge.",[11,669,670],{},"The distinction between resolved inputs and broad context matters. A packet should say which approved brief to use, not provide a folder and ask the agent to identify the authoritative version. It should name the relevant test command, not merely mention that tests exist. It should link a predecessor’s accepted artifact, not force the agent to search conversation history for the last apparently complete draft.",[11,672,673],{},"Context is useful when it reduces uncertainty. Beyond that point, it becomes another search surface.",[11,675,676],{},"The same principle applies to recovery. “Retry if needed” transfers diagnosis back to the agent. A targeted recovery hint might instead say that a truncated handoff can be retrieved with one exact read, that an unavailable credential requires returning the step with a specific reason, or that a validation failure should go back to the producing step with the failed criterion attached.",[11,678,679],{},"A good recovery path narrows the next decision without pretending every failure can be anticipated.",[24,681,683],{"id":682},"authority-should-match-the-active-step","Authority should match the active step",[11,685,686],{},"Instructions alone do not define what an agent can do. The executable step also needs an authority boundary.",[11,688,689],{},"A practical model grants tools and resources when the step becomes active, scoped to the operations and inputs required for that step. Completion, cancellation, or release of the step ends that authority. The agent does not need ambient access to every workflow capability, and it should not have to discover whether a documented operation is actually permitted.",[11,691,692],{},"The relationship should be legible: purpose leads to a permitted action, the action produces required evidence, and the evidence has a defined handoff.",[11,694,695],{},"If any link is missing, the agent must guess. If tools are described but unavailable, it enters recovery before substantive work begins. If broad tools are available without a step-level purpose, the system invites drift.",[11,697,698],{},"Approval can be part of this boundary when risk or organizational policy calls for it, but approval is not inherently required for every step. The important property is that authority is explicit, scoped, and legible to the acting agent.",[24,700,702],{"id":701},"review-feedback-does-not-own-delivery","Review feedback does not own delivery",[11,704,705],{},"Independent review is often represented as a simple gate after production. In operation, the reviewer also needs a claimed-step packet: the artifact under review, the governing criteria, relevant evidence, permitted verification tools, and a structured way to report findings.",[11,707,708],{},"An independent reviewer still needs the relevant context. It evaluates the work against declared criteria rather than inheriting the producer’s conclusions as facts.",[11,710,711],{},"The orchestrator has a different responsibility: adjudicating what the findings are allowed to change. A supported blocker can stop progression. A specific repair can return to the producing step. A preference can remain advice. An unsupported or out-of-scope finding should not expand the delivery.",[11,713,714],{},"This is a feedback gate, but not a ritual approval step. It protects the agreed goal from review-driven drift while preserving independent scrutiny where it matters.",[24,716,718],{"id":717},"a-bounded-observation-from-one-stackos-replay","A bounded observation from one StackOS replay",[11,720,721],{},"One recent StackOS cold-start replay provides a small implementation example. It should be read as first-party operating evidence, not as a productivity benchmark or universal proof.",[11,723,724],{},"During that replay, a draft specialist reported that it had used no tools and made no guesses while completing its assigned work. That report records the agent’s operating behavior alongside the artifact. We did not run a comparative test to isolate why it needed no additional discovery.",[11,726,727],{},"Another step received a truncated dependency handoff. Instead of searching broadly, the agent followed the packet’s targeted recovery hint and retrieved the one complete prior-step result it needed. Near the end of the run, the final stopping rule kept the active agent from starting another workflow, changing the workflow setup, or creating unrelated content after the requested outcome had been reached.",[11,729,730],{},"The run still contained ordinary latency and friction. Agents had to process context, produce outputs, and move through orchestration boundaries. One fresh subagent missed its bounded execution window. The observation establishes neither lower overhead nor gains across models or workflows.",[11,732,733],{},"Within that boundary, the replay records three useful behaviors: one specialist reported no guessing, targeted recovery constrained the response to a partial handoff, and an explicit stopping rule limited drift.",[24,735,737],{"id":736},"the-vague-cold-start-test","The vague cold-start test",[11,739,740],{},"A direct way to inspect agent experience is to remove accumulated familiarity.",[11,742,743],{},"Give a fresh agent the kind of vague request a real operator might provide. Do not give it the workflow key, the design rationale, or a warm context window containing earlier exploration. Then observe whether the system helps it discover the right workflow and whether the claimed step answers these questions:",[273,745,746,749,752,755,758,761,764],{},[82,747,748],{},"What is the first valid action?",[82,750,751],{},"Which exact inputs should be used?",[82,753,754],{},"Which tools are permitted and available?",[82,756,757],{},"What observable criteria define completion?",[82,759,760],{},"What must be produced, and where does it go?",[82,762,763],{},"What should happen if the expected path fails?",[82,765,766],{},"When must the agent stop and return control?",[11,768,769],{},"The test exposes friction when the agent must search broadly for policy, infer which artifact is authoritative, guess whether it has permission, recreate prior decisions, or invent a recovery strategy. Those behaviors may still produce a successful result, but they reveal orchestration work leaking into the execution step.",[11,771,772,773,776],{},"Teams can apply the same test while following a practical guide to ",[53,774,775],{"href":127},"building an AI agent workflow",". For each step, record the agent’s first action, unresolved questions, searches, inferred assumptions, unavailable tools, recovery attempts, and work performed after the completion condition. The result is a friction map grounded in behavior rather than a subjective rating.",[11,778,779],{},"Some ambiguity belongs to the work, some discovery is intentional, and some failures require judgment. The design target is narrower: stop making each fresh agent reconstruct decisions that the system has already made.",[11,781,782],{},"Orchestration determines what should happen next. Agent experience determines whether the next agent can actually do it.",{"title":300,"searchDepth":301,"depth":301,"links":784},[785,786,787,788,789,790],{"id":570,"depth":301,"text":571},{"id":594,"depth":301,"text":595},{"id":682,"depth":301,"text":683},{"id":701,"depth":301,"text":702},{"id":717,"depth":301,"text":718},{"id":736,"depth":301,"text":737},"AI agent orchestration is experienced one claimed step at a time. Here is how context, authority, tools, recovery, and stopping rules shape whether an agent can execute.",{},"\u002Farticles\u002Fai-agent-experience","2026-07-11","9 min read",[320,321,797],"branding-voice-reviewer",[799,325,323],"ai-agent-vs-workflow-vs-orchestrator",[327,328],"Understand how to design the operating experience inside AI agent orchestration",{"title":540,"description":791},"articles\u002Fai-agent-experience",[805,263,806],"AI agent orchestration","workflow design","uV1oEIIU0bekOi0eZ7egVtz64SWH2q_6b4myaRa7sf0",{"id":809,"title":810,"author":6,"body":811,"category":309,"description":1094,"extension":311,"featured":312,"heroImage":313,"meta":1095,"navigation":312,"path":1096,"publishedAt":794,"readingTime":795,"relatedAgents":1097,"relatedArticles":1099,"relatedWorkflows":1100,"searchIntent":1101,"seo":1102,"stem":1103,"topics":1104,"updatedAt":794,"visual":1106,"__hash__":1107},"articles\u002Farticles\u002Fhow-to-build-ai-agent-workflow.md","How to build an AI agent workflow: start with the problem",{"type":8,"value":812,"toc":1084},[813,816,819,823,830,834,837,840,860,863,866,872,876,879,885,891,897,903,906,910,913,916,919,925,928,931,935,938,941,944,950,954,957,960,1001,1004,1007,1010,1014,1017,1020,1023,1026,1029,1033,1036,1039,1042,1045,1049,1052,1081],[11,814,815],{},"Start a useful AI agent workflow with an operational problem: something that should move from an uncertain starting state to a useful, verifiable outcome. Models, prompt libraries, and agent diagrams come later.",[11,817,818],{},"That distinction changes the design. A chain of prompts describes what the model should say next. An operational contract describes what the system is allowed to do, what state it must preserve, how progress is evaluated, and when the job is finished.",[820,821],"article-workflow-visual",{"title":822,"workflow":327},"Design backward from a useful outcome",[11,824,825,826,829],{},"This is also what separates a workflow from adjacent concepts. An ",[53,827,828],{"href":565},"agent, workflow, and orchestrator"," can all involve model reasoning, but they carry different responsibilities. The workflow defines the operating boundary. The orchestrator decides how to advance within it. Agents perform bounded work.",[24,831,833],{"id":832},"why-prompt-chains-become-fragile","Why prompt chains become fragile",[11,835,836],{},"Prompt chains often look convincing in a prototype. One prompt gathers information, another drafts, and a third reviews. Each stage passes text to the next.",[11,838,839],{},"The fragility appears when the input is incomplete, a tool fails, a reviewer finds a real defect, or the work resumes after interruption. The chain has no reliable answer to questions such as:",[79,841,842,845,848,851,854,857],{},[82,843,844],{},"Which facts were verified, and where did they come from?",[82,846,847],{},"Is a review comment a blocker, a repair, or an unsupported preference?",[82,849,850],{},"Can a failed step be retried without repeating side effects?",[82,852,853],{},"What remains unfinished?",[82,855,856],{},"Has the requested outcome already been reached?",[82,858,859],{},"Which instructions still apply after several rounds of generated text?",[11,861,862],{},"Prompting alone does not create durable state or explicit control flow for these questions. A larger context window can hide that gap without closing it.",[11,864,865],{},"A prompt chain also tends to mix reasoning, state, and control flow. The model is expected to remember prior decisions, infer the current phase, choose tools, preserve constraints, and decide whether to stop. Small ambiguities accumulate. Later steps inherit summaries of summaries rather than a stable account of the job.",[11,867,868,869,871],{},"Representing those responsibilities explicitly makes an ",[53,870,587],{"href":586}," easier to inspect when something changes.",[24,873,875],{"id":874},"define-the-job-before-defining-the-agents","Define the job before defining the agents",[11,877,878],{},"Start with four descriptions: the problem, the useful outcome, the operator path, and the agent path.",[11,880,259,881,884],{},[607,882,883],{},"problem"," describes the operational gap. “We need a five-agent research system” is an implementation preference. “An editor cannot tell which claims in a draft are supported by the supplied sources” is a problem.",[11,886,259,887,890],{},[607,888,889],{},"useful outcome"," is the state in which that problem has been resolved. It should be inspectable. For the example above, the outcome might be a publishable draft whose material claims are linked to evidence, with unresolved claims clearly identified.",[11,892,259,893,896],{},[607,894,895],{},"operator path"," describes what a person initiating or supervising the workflow must do. What do they provide? Which choices can only they make? What can they inspect or revise? If the operator must repeatedly reconstruct hidden workflow state from chat history, the contract is incomplete.",[11,898,259,899,902],{},[607,900,901],{},"agent path"," describes how the system turns the initial inputs into the outcome. It names the required stages, dependencies, tools, state transitions, and recovery behavior. It should not assume that the agent will infer the intended route from a vague goal.",[11,904,905],{},"The two paths should meet at explicit interaction points, but they do not need a mandatory approval after every step. Some workflows can proceed automatically within a narrow boundary. Others need a decision when evidence conflicts, scope changes, or an external side effect is about to occur. The contract should reflect the actual risk rather than adding ceremonial checkpoints.",[24,907,909],{"id":908},"work-backward-from-a-terminal-condition","Work backward from a terminal condition",[11,911,912],{},"A workflow needs a definition of done that can survive imperfect execution.",[11,914,915],{},"“Produce a good article” is not a terminal condition. Neither is “continue until the reviewer is satisfied.” Both delegate completion to an unbounded judgment.",[11,917,918],{},"A stronger terminal condition combines observable state with acceptance criteria. For example:",[920,921,922],"blockquote",{},[11,923,924],{},"The draft exists, required sections are present, material claims have supporting evidence or an explicit unresolved status, blocking review findings are repaired, and the output passes the sanitization checks.",[11,926,927],{},"This gives the orchestrator something concrete to evaluate. It also prevents the workflow from looping because a reviewer can always imagine another improvement.",[11,929,930],{},"Terminal conditions should distinguish failure from incompleteness. Missing credentials, unavailable evidence, and contradictory operator requirements may prevent completion. Those states should produce structured recovery information: what failed, what was preserved, and what action could unblock the run.",[24,932,934],{"id":933},"keep-durable-state-outside-the-conversation","Keep durable state outside the conversation",[11,936,937],{},"Conversation context is useful working memory. It is a poor system of record.",[11,939,940],{},"Durable workflow state should capture the facts needed to resume or audit the job: inputs, source references, decisions, step status, outputs, findings, retries, and unresolved issues. Generated prose can be part of that state, but it should not be the only place where the workflow records what happened.",[11,942,943],{},"While building StackOS, we have treated this separation as an authoring constraint. The workflow contracts we are refining represent run state, step boundaries, grants, findings, and outputs independently from an agent’s conversational context. That is implementation experience, not proof that every workflow needs the same storage model. The useful principle is narrower: state required for correct continuation should not depend on a model reconstructing it from dialogue.",[11,945,946,947,949],{},"Durable state changes the ",[53,948,263],{"href":262},". An agent entering halfway through a run can inspect the current state instead of performing archaeology on a transcript.",[24,951,953],{"id":952},"make-every-step-an-explicit-packet","Make every step an explicit packet",[11,955,956],{},"A step should arrive as a bounded packet of work. A role name followed by the entire project history leaves the agent to reconstruct the real assignment.",[11,958,959],{},"A useful step packet contains:",[79,961,962,967,973,978,983,989,995],{},[82,963,964,966],{},[607,965,609],{}," why the step exists and what downstream decision it supports.",[82,968,969,972],{},[607,970,971],{},"Inputs:"," the artifacts, references, and state the step may rely on.",[82,974,975,977],{},[607,976,639],{}," the relevant constraints without unrelated run history.",[82,979,980,982],{},[607,981,627],{}," the operations available for this step, including their scope.",[82,984,985,988],{},[607,986,987],{},"Expected outputs:"," the artifact or state change the step must produce.",[82,990,991,994],{},[607,992,993],{},"Criteria:"," the checks that determine whether the output is acceptable.",[82,996,997,1000],{},[607,998,999],{},"Recovery:"," how to report missing inputs, tool failures, ambiguity, or partial work.",[11,1002,1003],{},"Consider a claim-review step. Its purpose is not to “improve the draft.” It is to identify material claims, compare them with allowed evidence, and emit structured findings. Its tools might permit reading sources and recording findings but not rewriting the article. Its output distinguishes supported claims, unsupported claims, and cases where the available evidence is inconclusive.",[11,1005,1006],{},"That packet gives the reviewer enough freedom to reason without giving it ownership of the whole delivery. It also makes failures local. If evidence is missing, the workflow can repair that dependency rather than restart content production.",[11,1008,1009],{},"Exact tools matter because capability is part of the contract. An instruction such as “do not publish” is weaker than a step that has no publishing operation available. Tool boundaries turn behavioral expectations into operating constraints.",[24,1011,1013],{"id":1012},"give-one-orchestrator-ownership-of-progression","Give one orchestrator ownership of progression",[11,1015,1016],{},"A practical default is one reasoning orchestrator that owns progression: inspecting state, selecting the next eligible step, evaluating outputs, and checking the terminal condition. Add a specialist when the task needs a distinct context, tool boundary, or evaluation discipline—not simply because the brief contains several kinds of work.",[11,1018,1019],{},"In a content-production workflow we have been authoring for StackOS, one orchestrator coordinates bounded evidence, writing, claim, voice, and sanitization specialists. Each specialist has a different job and output contract. None independently decides that the entire article is complete.",[11,1021,1022],{},"The orchestrator also acts as the feedback gatekeeper. Review findings are classified before they affect delivery. A supported blocker can reopen an earlier step. A specific, valid repair can become bounded follow-up work. An unsupported preference or a finding outside the agreed scope does not silently expand the job.",[11,1024,1025],{},"This classification is designed to prevent a familiar multi-agent failure mode: every reviewer becomes a new source of authority. Without classification, one agent’s stylistic suggestion can override the original brief, trigger unnecessary rewrites, and create another review cycle. Feedback should change delivery only when the workflow contract says that kind of finding matters.",[11,1027,1028],{},"This is not a universal claim that every system needs exactly one orchestrator. It is a practical default for workflows where several bounded tasks contribute to one outcome. Additional reasoning authorities should have a clear ownership boundary, not merely a different persona.",[24,1030,1032],{"id":1031},"verify-the-cold-start","Verify the cold start",[11,1034,1035],{},"A workflow that succeeds only when its designer supplies unstated context is not finished.",[11,1037,1038],{},"Cold-start verification gives a fresh agent the kind of vague request an actual operator might provide and observes what happens. Does the agent locate the workflow? Does it inspect state and requirements? Does it ask for a genuinely missing choice? Or does it guess the project, invent inputs, and begin producing output?",[11,1040,1041],{},"While refining our StackOS workflow guidance, we use fresh-agent scenarios to expose these gaps. We are testing whether the operating contract leads an unfamiliar agent toward the intended path, not whether the model can improvise a plausible response.",[11,1043,1044],{},"A good cold start should make the safe next action easier than guessing.",[24,1046,1048],{"id":1047},"a-compact-workflow-design-test","A compact workflow design test",[11,1050,1051],{},"Before adding another prompt or specialist, test the workflow with a short sequence of questions:",[273,1053,1054,1057,1060,1063,1066,1069,1072,1075,1078],{},[82,1055,1056],{},"What operational problem is being resolved?",[82,1058,1059],{},"What observable state counts as a useful outcome?",[82,1061,1062],{},"What does the operator provide, decide, and receive?",[82,1064,1065],{},"What path may the agent take, and which actions are outside its boundary?",[82,1067,1068],{},"Where does durable state live?",[82,1070,1071],{},"Does each step have explicit inputs, tools, outputs, criteria, and recovery?",[82,1073,1074],{},"Who evaluates findings and decides whether they change the run?",[82,1076,1077],{},"Can a fresh agent find the path without private context?",[82,1079,1080],{},"Can the workflow stop deterministically?",[11,1082,1083],{},"If the answers are vague, a more elaborate agent topology can make the ambiguity harder to see. Start with the problem, define the contract, and let the agents occupy only the boundaries the work actually requires.",{"title":300,"searchDepth":301,"depth":301,"links":1085},[1086,1087,1088,1089,1090,1091,1092,1093],{"id":832,"depth":301,"text":833},{"id":874,"depth":301,"text":875},{"id":908,"depth":301,"text":909},{"id":933,"depth":301,"text":934},{"id":952,"depth":301,"text":953},{"id":1012,"depth":301,"text":1013},{"id":1031,"depth":301,"text":1032},{"id":1047,"depth":301,"text":1048},"A practical way to design AI agent workflows from the outcome backward, with explicit state, step contracts, feedback boundaries, and cold-start verification.",{},"\u002Farticles\u002Fhow-to-build-ai-agent-workflow",[1098,319,321],"stackos-workflow-workflow-author",[799,325,324],[327,328],"Learn how to build an AI agent workflow from the problem and outcome backward",{"title":810,"description":1094},"articles\u002Fhow-to-build-ai-agent-workflow",[1105,335,334],"AI agent workflows","workflow","I59u29X8EdoopdGoZws9pKiDaHQtKY8L6q2TQAeeyeU",{"id":1109,"title":1110,"author":6,"body":1111,"category":309,"description":1231,"extension":311,"featured":312,"heroImage":313,"meta":1232,"navigation":312,"path":1233,"publishedAt":1234,"readingTime":1235,"relatedAgents":1236,"relatedArticles":1237,"relatedWorkflows":1239,"searchIntent":1240,"seo":1241,"stem":1242,"topics":1243,"updatedAt":1234,"visual":552,"__hash__":1247},"articles\u002Farticles\u002Fai-agent-vs-workflow-vs-orchestrator.md","AI agent vs. workflow vs. orchestrator: what is the difference?",{"type":8,"value":1112,"toc":1223},[1113,1116,1119,1123,1127,1130,1133,1141,1145,1148,1151,1154,1158,1161,1164,1171,1175,1178,1197,1200,1204,1207,1210,1214,1217,1220],[11,1114,1115],{},"An AI agent performs a focused role. A workflow defines the stages and rules of the work. An orchestrator coordinates the workflow, choosing the right agents, context, and tools as the job changes.",[11,1117,1118],{},"The terms are often used interchangeably, but separating them makes AI-powered work much easier to design and manage.",[18,1120],{"caption":1121,"mode":552,"title":1122},"Agents perform focused roles. The workflow defines the path. The orchestrator keeps the complete job moving.","One job, three different responsibilities",[24,1124,1126],{"id":1125},"what-is-an-ai-agent","What is an AI agent?",[11,1128,1129],{},"An AI agent is a model operating with a role, instructions, context, and tools. Its role should be focused enough that its decisions can be understood and checked.",[11,1131,1132],{},"A content workflow might use an evidence curator, a writer, and a claim reviewer. A delivery workflow might use a designer, implementer, tester, and reviewer. These specialists can use the same underlying model while having different responsibilities and boundaries.",[11,1134,1135,1136,1140],{},"See the ",[53,1137,1139],{"href":1138},"\u002Flibrary\u002Fagents","agent library"," for examples of focused roles used by StackOS workflows.",[24,1142,1144],{"id":1143},"what-is-a-workflow","What is a workflow?",[11,1146,1147],{},"A workflow is the repeatable shape of the work. It defines the stages that matter, what each stage needs, which stages depend on others, where approval is required, and what a successful result looks like.",[11,1149,1150],{},"The workflow is reusable, but each job is specific. “Produce a campaign” may always require research, planning, creation, and review, while the actual steps expand based on the channels, source material, and goals in the request.",[11,1152,1153],{},"This is why a workflow is more useful than a long prompt. It holds the relationships and state around the work, not only instructions for one response.",[24,1155,1157],{"id":1156},"what-is-an-orchestrator","What is an orchestrator?",[11,1159,1160],{},"An orchestrator is the coordinating role for the complete job. It understands the workflow, assembles the right context, brings in specialists at the right time, protects approval boundaries, and keeps the plan consistent as new information appears.",[11,1162,1163],{},"It does not need to perform every task itself. In fact, strong coordination usually separates creation from review so the same role is not judging its own work.",[11,1165,259,1166,1170],{},[53,1167,1169],{"href":1168},"\u002Flibrary\u002Forchestrators","orchestrator library"," shows how StackOS coordinates content, engineering, and marketing work.",[24,1172,1174],{"id":1173},"how-do-they-work-together","How do they work together?",[11,1176,1177],{},"Think of producing a campaign:",[79,1179,1180,1185,1191],{},[82,1181,259,1182,1184],{},[607,1183,1106],{}," says research must finish before the angle is selected, and review must finish before publishing.",[82,1186,259,1187,1190],{},[607,1188,1189],{},"agents"," research the evidence, shape the plan, write the assets, and review the claims.",[82,1192,259,1193,1196],{},[607,1194,1195],{},"orchestrator"," makes sure each specialist receives the right input, updates the plan when the campaign changes, and stops at the approval point before anything goes live.",[11,1198,1199],{},"StackOS keeps these parts connected to the actual apps where the work happens. The AI client remains the place where you ask and direct. StackOS becomes the place where the complete job stays organized.",[24,1201,1203],{"id":1202},"do-you-need-multiple-models","Do you need multiple models?",[11,1205,1206],{},"No. Different agents are roles, not necessarily different models. One model can take several roles at different stages, or a team can choose different models for different strengths.",[11,1208,1209],{},"The more important separation is responsibility. A focused research role should preserve sources. A review role should be able to challenge the draft. An orchestrator should coordinate without quietly bypassing approvals.",[24,1211,1213],{"id":1212},"which-part-should-you-design-first","Which part should you design first?",[11,1215,1216],{},"Start with the outcome and the workflow. Ask what must be true before the work is considered done, where mistakes are expensive, and which stages depend on earlier evidence.",[11,1218,1219],{},"Then assign focused agents to those responsibilities. Add an orchestrator when the work spans several stages, specialists, tools, or sessions.",[11,1221,1222],{},"This order keeps the system grounded in real work. You are not collecting agents because they sound impressive; you are giving each necessary part of the job a clear owner.",{"title":300,"searchDepth":301,"depth":301,"links":1224},[1225,1226,1227,1228,1229,1230],{"id":1125,"depth":301,"text":1126},{"id":1143,"depth":301,"text":1144},{"id":1156,"depth":301,"text":1157},{"id":1173,"depth":301,"text":1174},{"id":1202,"depth":301,"text":1203},{"id":1212,"depth":301,"text":1213},"Agents perform focused roles, workflows define how work moves, and orchestrators coordinate the whole job. Here is how the three fit together.",{},"\u002Farticles\u002Fai-agent-vs-workflow-vs-orchestrator","2026-07-09","6 min read",[320,321],[325,1238],"use-codex-claude-gemini-with-existing-tools",[327,328],"Compare AI agents, workflows, and orchestrators in plain language",{"title":1110,"description":1231},"articles\u002Fai-agent-vs-workflow-vs-orchestrator",[1244,1245,1246],"AI agents","orchestrators","agentic workflows","5k8G3D6QZtrxhpl399L2aCDCRmqFC_hRSslJnk7_cVo",{"id":1249,"title":1250,"author":6,"body":1251,"category":1360,"description":1361,"extension":311,"featured":1362,"heroImage":313,"meta":1363,"navigation":312,"path":1364,"publishedAt":1234,"readingTime":1365,"relatedAgents":1366,"relatedArticles":1368,"relatedWorkflows":1369,"searchIntent":1370,"seo":1371,"stem":1372,"topics":1373,"updatedAt":1234,"visual":1262,"__hash__":1377},"articles\u002Farticles\u002Fhow-ai-agents-use-accounts-safely.md","How can AI agents use business accounts without seeing the login?",{"type":8,"value":1252,"toc":1353},[1253,1256,1259,1264,1268,1271,1274,1278,1281,1284,1288,1291,1294,1308,1311,1315,1318,1321,1325,1328,1342,1345],[11,1254,1255],{},"AI agents can use business accounts without receiving the password or API key. The secret stays inside a trusted local process, while the model receives only a safe account reference and permission to request a specific action.",[11,1257,1258],{},"This boundary matters because an AI needs the ability to work, not a copy of every login.",[18,1260],{"caption":1261,"mode":1262,"title":1263},"Private account details stay inside the local StackOS process while the AI receives only the safe context and result it needs.","security","The model directs. StackOS performs the approved action.",[24,1265,1267],{"id":1266},"what-should-the-model-receive","What should the model receive?",[11,1269,1270],{},"The model needs enough information to make a good decision: which account is available, what it can do, whether it is ready, and what approval is required.",[11,1272,1273],{},"It does not need the raw token, password, or private key. StackOS provides a safe reference that identifies the connection without exposing the secret.",[24,1275,1277],{"id":1276},"where-does-the-secret-stay","Where does the secret stay?",[11,1279,1280],{},"StackOS runs locally on the user’s Mac. Connected credentials are resolved inside that local process only when an explicit action is being performed.",[11,1282,1283],{},"The model asks for an intent-level action—such as publishing an approved post to a selected site. StackOS checks the workflow permission, selected account, action contract, and approval before making the call.",[24,1285,1287],{"id":1286},"what-prevents-an-agent-from-doing-anything-it-wants","What prevents an agent from doing anything it wants?",[11,1289,1290],{},"Access is scoped to the work. A workflow stage receives only the tools allowed for that stage, and sensitive or costly actions can require approval.",[11,1292,1293],{},"That creates several useful boundaries:",[79,1295,1296,1299,1302,1305],{},[82,1297,1298],{},"A research stage can read approved information without receiving publishing access.",[82,1300,1301],{},"A writer can prepare content without being allowed to send it.",[82,1303,1304],{},"A review stage can inspect the result without changing external systems.",[82,1306,1307],{},"A publishing stage can use one approved account for one approved action.",[11,1309,1310],{},"Each action is recorded with its result so a person can understand what happened later.",[24,1312,1314],{"id":1313},"is-local-software-enough-by-itself","Is local software enough by itself?",[11,1316,1317],{},"Running locally reduces how far secrets have to travel, but location is only part of the design. Safe agent access also needs narrow permissions, explicit actions, approval gates, useful error handling, and a complete history.",[11,1319,1320],{},"The goal is not to claim that an AI agent can never make a mistake. The goal is to make its authority understandable, bounded, and reviewable.",[24,1322,1324],{"id":1323},"what-should-teams-ask-before-connecting-an-account","What should teams ask before connecting an account?",[11,1326,1327],{},"Ask four questions:",[273,1329,1330,1333,1336,1339],{},[82,1331,1332],{},"Which exact actions does this connection allow?",[82,1334,1335],{},"Which workflow stages can request those actions?",[82,1337,1338],{},"Which actions require a person to approve them?",[82,1340,1341],{},"What evidence will be recorded after the action runs?",[11,1343,1344],{},"If those answers are unclear, the connection is too broad.",[11,1346,1347,1348,1352],{},"StackOS is built around this local trust boundary. The ",[53,1349,1351],{"href":1350},"\u002Flibrary\u002Fworkflows","workflow library"," shows where connected actions fit inside complete, visible work rather than appearing as isolated tool calls.",{"title":300,"searchDepth":301,"depth":301,"links":1354},[1355,1356,1357,1358,1359],{"id":1266,"depth":301,"text":1267},{"id":1276,"depth":301,"text":1277},{"id":1286,"depth":301,"text":1287},{"id":1313,"depth":301,"text":1314},{"id":1323,"depth":301,"text":1324},"Security","A local action layer can keep private credentials away from the model while still letting AI perform explicit, approved work in connected apps.",false,{},"\u002Farticles\u002Fhow-ai-agents-use-accounts-safely","5 min read",[1367],"branding-sanitization-reviewer",[1238,325],[328,327],"Understand how AI agents can use connected accounts without receiving credentials",{"title":1250,"description":1361},"articles\u002Fhow-ai-agents-use-accounts-safely",[1374,1375,1376],"AI agent security","credentials","local software","ct8Jul2KFiiCNRPoR13www3awDGh2JJjamtceucT8_I",{"id":1379,"title":1380,"author":6,"body":1381,"category":1494,"description":1495,"extension":311,"featured":312,"heroImage":313,"meta":1496,"navigation":312,"path":1497,"publishedAt":1234,"readingTime":1235,"relatedAgents":1498,"relatedArticles":1499,"relatedWorkflows":1501,"searchIntent":1503,"seo":1504,"stem":1505,"topics":1506,"updatedAt":1234,"visual":21,"__hash__":1511},"articles\u002Farticles\u002Fuse-codex-claude-gemini-with-existing-tools.md","How to use Codex, Claude Code, or Gemini with the tools you already have",{"type":8,"value":1382,"toc":1486},[1383,1386,1389,1393,1397,1400,1403,1407,1410,1413,1416,1420,1423,1426,1430,1433,1436,1440,1443,1465,1468,1472,1475],[11,1384,1385],{},"You do not need to replace Codex, Claude Code, Gemini, or the business apps your team already uses. Connect the AI client to StackOS, connect StackOS to your approved tools, and keep working from the interface you prefer.",[11,1387,1388],{},"This creates a clean division of responsibility: the AI understands and directs the work; StackOS organizes its plan, state, tool access, approvals, and history.",[18,1390],{"caption":1391,"mode":21,"title":1392},"StackOS connects the conversation to a visible plan and the approved tools that complete it.","Keep the AI client. Keep the apps.",[24,1394,1396],{"id":1395},"why-keep-the-ai-client-separate","Why keep the AI client separate?",[11,1398,1399],{},"AI clients improve quickly and people have different preferences. One person may work in Codex, another in Claude Code, and another through Gemini. Locking the whole operating process to one chat interface makes switching expensive and fragments the work.",[11,1401,1402],{},"A shared work layer lets the client change without losing the project’s workflows, connected apps, or history.",[24,1404,1406],{"id":1405},"what-does-the-connection-look-like","What does the connection look like?",[11,1408,1409],{},"The AI client connects to StackOS through a standard tool interface. When you make a request, the model can find the relevant workflow, adapt it to the request, and present the plan.",[11,1411,1412],{},"After approval, work moves step by step. When a stage needs an outside app—such as Slack, Shopify, WordPress, an ad platform, or an analytics tool—StackOS performs the specific approved action and returns a safe result to the AI.",[11,1414,1415],{},"Your private login stays in the local StackOS process. The model receives a safe reference and the result it needs, not the secret itself.",[24,1417,1419],{"id":1418},"does-stackos-replace-automation-tools","Does StackOS replace automation tools?",[11,1421,1422],{},"StackOS is designed to coordinate work across existing tools. It does not ask a content team to abandon WordPress, a commerce team to replace Shopify, or an engineering team to stop using GitHub.",[11,1424,1425],{},"It adds the missing continuity between the request and those tools: which workflow applies, what is currently ready, what needs review, what action was taken, and where the result belongs.",[24,1427,1429],{"id":1428},"what-happens-when-you-change-models","What happens when you change models?",[11,1431,1432],{},"The durable work remains in StackOS. A new compatible AI client can recover the project, current plan, completed steps, decisions, and next actions instead of relying on the memory of one chat thread.",[11,1434,1435],{},"That makes model choice a practical preference rather than an operating-system decision.",[24,1437,1439],{"id":1438},"a-simple-example","A simple example",[11,1441,1442],{},"Suppose you ask Codex to investigate customer feedback and prepare a fix.",[273,1444,1445,1453,1456,1459,1462],{},[82,1446,1447,1448,1452],{},"Codex identifies the request and opens the appropriate ",[53,1449,1451],{"href":1450},"\u002Flibrary\u002Fworkflows\u002Fcommunications-customer-feedback-intake","customer feedback workflow",".",[82,1454,1455],{},"StackOS gathers the approved conversation context and creates visible work.",[82,1457,1458],{},"The investigation uses the connected communication and project tools.",[82,1460,1461],{},"If delivery is needed, the result moves into a tracked delivery workflow with its dependencies intact.",[82,1463,1464],{},"You can review what happened, what remains, and the evidence behind the conclusion.",[11,1466,1467],{},"The same pattern works when the starting client is Claude Code or Gemini.",[24,1469,1471],{"id":1470},"what-do-you-need-to-get-started","What do you need to get started?",[11,1473,1474],{},"Install StackOS on your Mac, connect a supported AI client, and add the apps you want it to use. Start with one real workflow that matters to your team rather than trying to automate everything at once.",[11,1476,1477,1478,1480,1481,1485],{},"Browse the ",[53,1479,1351],{"href":1350}," to choose a starting point, or ",[53,1482,1484],{"href":1483},"\u002F#install","download StackOS for Mac"," and connect the client you already use.",{"title":300,"searchDepth":301,"depth":301,"links":1487},[1488,1489,1490,1491,1492,1493],{"id":1395,"depth":301,"text":1396},{"id":1405,"depth":301,"text":1406},{"id":1418,"depth":301,"text":1419},{"id":1428,"depth":301,"text":1429},{"id":1438,"depth":301,"text":1439},{"id":1470,"depth":301,"text":1471},"Getting started","Keep your preferred AI client and existing business apps. Add a shared work layer so requests become trackable plans with safe, explicit actions.",{},"\u002Farticles\u002Fuse-codex-claude-gemini-with-existing-tools",[],[325,1500],"how-ai-agents-use-accounts-safely",[1502,328],"communications-customer-feedback-intake","Learn how to connect an existing AI client to existing business tools",{"title":1380,"description":1495},"articles\u002Fuse-codex-claude-gemini-with-existing-tools",[1507,1508,1509,1510],"Codex","Claude Code","Gemini","AI tools","C43pDZZGuDV2XSUh3op_ayhHUJJ2UYL1TSE6kjNv3lI",{"id":1513,"title":1514,"author":6,"body":1515,"category":1674,"description":1675,"extension":311,"featured":312,"heroImage":313,"meta":1676,"navigation":312,"path":1677,"publishedAt":1234,"readingTime":1678,"relatedAgents":1679,"relatedArticles":1681,"relatedWorkflows":1682,"searchIntent":1684,"seo":1685,"stem":1686,"topics":1687,"updatedAt":1234,"visual":1106,"__hash__":1689},"articles\u002Farticles\u002Fwhat-is-an-agentic-workflow.md","What is an agentic workflow? A practical guide to AI-powered work",{"type":8,"value":1516,"toc":1666},[1517,1520,1531,1534,1538,1541,1544,1547,1579,1583,1586,1594,1597,1600,1604,1607,1610,1613,1617,1620,1640,1646,1650,1653,1656,1660,1663],[11,1518,1519],{},"An agentic workflow is a structured way for an AI agent to move from a goal to a completed, checked result. It gives the AI a sequence of steps, the right context and tools for each step, clear approval points, and a record of what happened.",[11,1521,1522,1523,1527,1528,1530],{},"The important word is not ",[1524,1525,1526],"em",{},"AI",". It is ",[1524,1529,1106],{},". A capable model can reason about a request, but real work also needs continuity: what is ready, what is waiting, who approved a decision, which app was changed, and what evidence proves the result.",[820,1532],{"title":1533,"workflow":327},"A complete content workflow, from request to published result",[24,1535,1537],{"id":1536},"what-makes-a-workflow-agentic","What makes a workflow agentic?",[11,1539,1540],{},"An ordinary automation follows a fixed rule: when this happens, do that. An agentic workflow can interpret the goal, adapt the plan to the situation, use the right specialist for each stage, and respond when new information changes the work.",[11,1542,1543],{},"The workflow still provides boundaries. The AI does not get a blank check. It works through named stages, uses approved connections, and pauses when a person needs to decide.",[11,1545,1546],{},"A useful agentic workflow usually includes five parts:",[273,1548,1549,1555,1561,1567,1573],{},[82,1550,1551,1554],{},[607,1552,1553],{},"A goal."," The outcome the person actually wants, written in business language.",[82,1556,1557,1560],{},[607,1558,1559],{},"A plan."," The steps, relationships, checks, and approval points needed for this particular request.",[82,1562,1563,1566],{},[607,1564,1565],{},"Specialists."," Focused AI roles for research, creation, review, or other distinct responsibilities.",[82,1568,1569,1572],{},[607,1570,1571],{},"Connected tools."," The apps and accounts where the work already happens.",[82,1574,1575,1578],{},[607,1576,1577],{},"A durable work record."," Status, decisions, outputs, dependencies, and evidence that remain available after the chat ends.",[24,1580,1582],{"id":1581},"what-happens-from-request-to-result","What happens from request to result?",[11,1584,1585],{},"Imagine a content leader asks Claude Code to turn customer research into a four-week campaign. The conversation is only the starting point.",[11,1587,1588,1589,1593],{},"StackOS recognizes the matching ",[53,1590,1592],{"href":1591},"\u002Flibrary\u002Fworkflows\u002Fbranding-content-production","content production workflow",", adds the right stages for the request, and presents the complete plan before work begins. A research specialist gathers the source material. A strategist shapes the angle and channel plan. A writer creates the main piece. Separate reviewers check claims, voice, and disclosure risk. Approved work then moves to the connected publishing tools.",[11,1595,1596],{},"Each stage changes state as it moves from waiting to working, review, and done. If the research is incomplete, later work stays blocked. If a reviewer finds an unsupported claim, the piece returns to the right stage with the reason attached.",[11,1598,1599],{},"That visible state is what turns a promising AI conversation into dependable work.",[24,1601,1603],{"id":1602},"how-is-this-different-from-a-chatbot","How is this different from a chatbot?",[11,1605,1606],{},"A chatbot mainly responds inside a conversation. It may produce an excellent answer, but the person still has to carry the result into other apps, remember what remains, and explain the context again next time.",[11,1608,1609],{},"An agentic workflow gives the conversation somewhere to go. The request becomes organized work with owners, dependencies, connected actions, approvals, and proof.",[11,1611,1612],{},"You can still use the AI interface you prefer. StackOS works with Codex, Claude Code, Gemini, and other tool-using AI clients; it adds the shared work layer around them.",[24,1614,1616],{"id":1615},"where-can-agentic-workflows-be-used","Where can agentic workflows be used?",[11,1618,1619],{},"The pattern is not limited to software development. Any repeatable outcome with judgment, multiple steps, connected apps, or approval risk can benefit.",[79,1621,1622,1625,1628,1631,1634,1637],{},[82,1623,1624],{},"Engineering teams can move from a feature request through design, delivery, testing, and review.",[82,1626,1627],{},"Content teams can research, draft, fact-check, create visuals, and publish.",[82,1629,1630],{},"Finance teams can collect data, investigate a variance, prepare a recommendation, and route it for approval.",[82,1632,1633],{},"Commerce teams can update product content, review merchandising changes, and coordinate Shopify operations.",[82,1635,1636],{},"Sales teams can research accounts, enrich leads, prepare outreach, and keep the CRM accurate.",[82,1638,1639],{},"Support teams can investigate an issue and hand confirmed work to the right delivery team.",[11,1641,1642,1643,1645],{},"Explore the ",[53,1644,1351],{"href":1350}," to see how the same structure adapts across different kinds of work.",[24,1647,1649],{"id":1648},"when-should-you-use-one","When should you use one?",[11,1651,1652],{},"Use an agentic workflow when the work needs more than one answer. Strong signals include handoffs between people or apps, steps that depend on earlier results, sensitive actions, required review, and work that may continue across multiple sessions.",[11,1654,1655],{},"For a quick question or one-off draft, a normal conversation may be enough. For work that must finish correctly and remain understandable later, a workflow gives the AI—and the team—the structure it needs.",[24,1657,1659],{"id":1658},"the-shortest-useful-definition","The shortest useful definition",[11,1661,1662],{},"An agentic workflow is a goal-driven plan that lets AI complete multi-step work through connected tools while keeping progress, approvals, and results visible.",[11,1664,1665],{},"That is the practical promise: keep the AI tool you already like, and give its work a reliable path from request to result.",{"title":300,"searchDepth":301,"depth":301,"links":1667},[1668,1669,1670,1671,1672,1673],{"id":1536,"depth":301,"text":1537},{"id":1581,"depth":301,"text":1582},{"id":1602,"depth":301,"text":1603},{"id":1615,"depth":301,"text":1616},{"id":1648,"depth":301,"text":1649},{"id":1658,"depth":301,"text":1659},"Agentic workflows","An agentic workflow turns a goal into visible steps that AI can plan, complete, check, and hand off across the tools your team already uses.",{},"\u002Farticles\u002Fwhat-is-an-agentic-workflow","7 min read",[1680,320],"branding-evidence-curator",[799,1238],[328,327,1683],"marketing-campaign-production","Understand what an agentic workflow is, how it works, and when to use one",{"title":1514,"description":1675},"articles\u002Fwhat-is-an-agentic-workflow",[1246,1244,1688],"workflow automation","p3F0WaG5bUQyMS33NZdcvFDy40J9TTEmz_HtpsjB0Lw",1783893477533]