Where Should a Business Start with AI Agents? (2026)
Quick summary: Start with readiness out of 30 — below 16 out of 30, do not let it change orders — then one family, usually support reads. McKinsey found 62% experimenting and 23% scaling.
Key Takeaways
- Start with readiness out of 30 — below 16 out of 30, do not let it change orders — then one family, usually support reads
- McKinsey found 62% experimenting and 23% scaling
- McKinsey's State of AI 2025 put 62% of organizations at least experimenting with agents and 23% scaling in at least one function
- This is part 4 of AI Agents for Business
- Skip it when the total is below 16 out of 30
Table of Contents
Most businesses should start with one read-only agent on one workflow, after a readiness score, not with a fleet and a keynote. McKinsey’s State of AI 2025 put 62% of organizations at least experimenting with agents and 23% scaling in at least one function. Experimenting is not a starting plan. A score and a single family is.
This is part 4 of AI Agents for Business. The interactive picker is Which eCommerce agent first?. The org-wide score is the readiness assessment.
The job. Score readiness, pick one family, ship week-one reads — not a supervisor swarm.
This week. Fill the readiness checklist with one ops lead and one engineer in the same hour.
A person still signs. Refunds, POs, and anything on the blast-radius veto — even after the score passes.
Skip it when the total is below 16 out of 30. Below that, do not let it change orders. Fix join keys and ownership first.
Reproduce this — Fill the checklist above. Score each row 0/1/2. Total /30. Then walk the decide tree. Interactive readiness: Agentic commerce readiness checker. Series folder:
ecommerce-ai-agents-series/.
FactualMinds is an AWS Select Tier Services Partner. We have no published AI-agent case studies. The decide tree and the checklist are the artifacts we can point at.
Our take: if the score is under 16 out of 30, start by fixing join keys, named APIs, and a human queue — not by picking a model. You delay the demo. You avoid a chatbot that refunds from a prompt.
The starting order we recommend
- Score readiness /30 — data, integration, process, governance, priority. Flagship write-up: readiness assessment.
- Pick one family — support, sales, operations, inventory, or knowledge — on the hub or the decide tree.
- Week-one reads only — especially support. See how agents reduce support work.
- One write later — after Policy, evals, and a named owner. Not a supervisor swarm.
Default first family (and when to override)
| Start here | When this is the right first move | When it is the wrong first move |
|---|---|---|
| Support | Tickets repeat; OMS API named; WISMO/policy is a visible slice | Tickets already closed by templates; leadership only wants auto-refund |
| Inventory | ATP is trustworthy; a buyer already owns POs | Inventory file is a weekly CSV with no join keys |
| Operations | Someone already writes a daily priority list | You want a warehouse rewrite |
| Sales | Contract price exists and is queryable | The PDP is the only price |
| Knowledge | Catalog IDs do not join to OMS/WMS | You want a generic research bot |
What broke — Teams treated a shopping-agent PDP/API score as org readiness and attached writes. Detection: no owner, no goldens, no escalation on the first bad
createReturn. Recovery: run the /30 checklist; stay on reads until 16+. That counter-case is published in the readiness post.
Named substitutes
- Score under 16 out of 30 → fix data and ownership; readiness post. No writes.
- Cannot choose among families → decide tree, not a five-agent program.
- Want cost math before a family → how to evaluate opportunities and the ROI calculator.
- Want the automation map → 15 automations in the field guide — not this executive note.
If you only do one thing
Open the decide tree after you have a /30 score. If you skip the score, you will pick the family that looked good in a vendor demo.
For your technical lead
On June 17, 2026, AgentCore Harness reached GA (What’s New). That date is a hosting fact. It is not a green light. Agents Classic is in maintenance for new customers after July 30, 2026.
First-party signals we reuse (not client KPIs) — Gateway ~180 ms → ~95 ms on a B2B CRM assistant (12 tools, ~8k turns/day) — Gateway post. Support-style AgentCore at 50K sessions/mo ~$791/mo platform + model (decision guide). Treat ~$791/mo as a platform cost floor to plan against, not savings.
What to do this week
- Fill the readiness checklist with one ops lead and one engineer in the same hour.
- If the total is under 16 out of 30, schedule data/API work. Do not book a write-capable build.
- If the total is 16+, walk the decide tree and open one family page.
- Write the week-one stop rule before anyone creates a harness.
What this post doesn’t cover
It does not replace the decide tree’s branching questions. It does not score shopping-agent PDP readiness — that is a different field-guide post. It does not recommend a model or a vendor bake-off. We have not published a new “first family” census; 16/30, 62%/23%, ~180→95 ms, and ~$791/mo are the published figures we reuse.
Primary next step: Which eCommerce agent first?.
AWS Cloud Architect & AI Expert
AWS-certified cloud architect and AI expert with deep expertise in cloud migrations, cost optimization, and generative AI on AWS.
