AI is changing who can attempt a job before most companies have changed who is responsible for the result. OpenAI's July research analyzed more than 800,000 messages from U.S. ChatGPT users and found that 43.5% of occupation-specific messages involved tasks associated with another occupation. The crossover was more pronounced in smaller organizations, where the person closest to a problem often handles it instead of handing it to a specialist.

That pattern already feels familiar in a building business. A project manager can draft marketing copy. A designer can troubleshoot a spreadsheet. An owner can review contract language. An estimator can build a client presentation. AI lowers the skill barrier around the first draft, but it does not transfer professional judgment, approval authority, or business accountability.

Task access is not decision authority

The dangerous assumption is that whoever can now produce an artifact can also approve or act on it. A project manager may use AI to compare subcontractor proposals without gaining authority to change the scope. A designer may draft a product substitution without owning purchasing risk. An administrator may summarize a contract without becoming legal counsel. The workflow must distinguish creation, review, approval, and execution.

Stanford HAI's worker study supports that distinction. In a survey of 1,500 workers across 104 occupations, respondents generally preferred collaborative AI and human oversight at critical points. They welcomed help with repetitive work, scheduling, file maintenance, and record correction, while trust, reliability, and loss of oversight remained major concerns.

Add a responsibility map to every AI workflow

  • Requester: the person who starts the job and defines the business need.
  • Source owner: the person responsible for confirming that drawings, scopes, quotes, policies, or records are current.
  • AI worker: the narrow workflow that researches, compares, drafts, calculates, or organizes the approved source packet.
  • Reviewer: the role qualified to check completeness, reasoning, calculations, citations, and job-specific judgment.
  • Approver: the person authorized to commit money, communicate externally, change scope, alter records, or accept risk.
  • Escalation owner: the person who resolves conflicts, missing sources, policy exceptions, and failed runs.

Keep those roles visible on the workflow card or run record. In a small company, one person may occupy several roles, but naming each role still matters. It forces the team to notice when the person producing the draft is also approving it without an independent check.

The handoff is part of the product

OpenAI's current production-agent guidance starts each deployment with a specific job, only the required knowledge and system access, explicit policies, approved actions, evaluations, and escalation rules. That is useful far beyond enterprise customer service. A remodeler's quote-comparison workflow should know when a document is missing, when totals disagree, when an alternate changes scope, and exactly who must take over.

A good handoff is structured, not conversational fog. It should state what the AI completed, which sources it used, what remains uncertain, what decision is needed, and which actions are still locked. The reviewer should not have to reconstruct the work from a long chat transcript.

Evaluate the boundary, not just the answer

Before reuse, test whether the workflow respects responsibility boundaries. Give it an outdated drawing, conflicting quote totals, an unsupported client request, a missing approval, and a task outside its assigned job. A useful evaluation checks whether the AI cites the right source, flags the conflict, routes the decision to the correct role, and refuses to take an unauthorized action.

Track reviewer corrections too. If the estimator repeatedly fixes exclusions, the source packet or comparison template needs work. If the owner repeatedly rewrites client commitments, the approval boundary is too late. The goal is not to prove the model is impressive. It is to make responsibility clearer and rework smaller each time the workflow runs.

Clear ownership is also better public proof

Google's current guidance for AI Overviews and AI Mode says there is no special AI-only schema or writing hack. Useful, original, technically clear content remains the foundation, and structured data should support what is visible. For Datum, that means public guidance should expose the operating method an AI summary cannot replace: the roles, checkpoints, source rules, and evaluation cases a building business can actually adopt.

Continue with the operating system

Sources Read

Next step, if this note maps to a problem on your desk: Private Training — a private working session for your team ($1,500+).

Related Field Notes