An agent becoming available does not establish that your business needs one. Treat a proposed workflow as an operating decision: what complete job will it support, what decisions remain with people, and who owns the result when the tool is unavailable?
Three questions before procurement
What specific job is it doing? Describe the trigger, permitted source information, accepted output, decision owner and exceptions. “Follow up approved quotes” is more bounded than “AI for the business”, but still requires permissions, timing rules and evidence of what happened. Writing a clear job description does not prove it is technically or commercially suitable.
What can it touch? Distinguish reading, drafting, updating records and communicating externally. Each needs its own authority. Inspect permissions and failure behaviour rather than assuming that a brand or a platform category supplies safe controls.
What happens when it is uncertain or fails? Specify a stop or review path, retained evidence, an accountable operator and recovery arrangements. A confidence score alone does not establish correctness. Include stale sources, conflicting instructions, unavailable systems and unexpected inputs in representative tests.
Compare complete work
Inspect the existing tools and operating method first. A checklist, clearer handoff or supported feature may resolve the constraint. Compare preparation, review, corrections and accepted output, including provider costs, maintenance and exceptions. A faster draft can create more downstream checking; measure through acceptance.
If a bounded test is justified, state its cases, acceptance criteria and human decisions before running it. Successful test cases do not prove all production inputs will work. Record failures and limits, and decide separately whether an installation is warranted.
Operating ownership matters
Ask who maintains access, checks updates, handles incidents and supplies usable records at handover. Those responsibilities remain necessary whether the tool is self-managed or contracted. Availability of agent services does not make adoption mandatory or guarantee reliable business infrastructure.
Current implementation routes
Reviewed 2 November 2026. Specialist advisory remains our primary focus, with a second implementation lane for other small and boutique businesses. A free 25-minute Fit Call can lead to a scoped Build or Connected Build, an optional Blueprint, a smaller check or a simpler route. Detailed work and installation are quoted separately; existing signed agreements retain their terms.
Compare a ChatGPT or Claude workspace with a custom build, or choose a first workflow using real work. The two service pages describe ChatGPT + Claude Implementation and phased AI Transformation.