How to choose your first AI agent: a 10-question scorecard.
The hardest part of AI in 2026 isn’t finding an agent. It’s choosing the first one, so it actually pays back. Score your options on these ten questions, and the answer usually picks itself.
- Start with the workflow, not the tool.
- Score ten questions from 0 to 2.
- 15 or more: build an integrated agent. Under 10: start with a self-serve tool.
The scorecard
| # | Question | 0 | 1 | 2 |
|---|---|---|---|---|
| 1 | How often does the task happen? | Monthly | Weekly | Daily |
| 2 | What does it cost today? | Unclear | Some hours | Clear hours or revenue lost |
| 3 | How costly is a mistake? | Low | Medium | High |
| 4 | Is it rules-based? | Mostly judgment | Mixed | Mostly rules |
| 5 | Does it live in your systems? | No | Partly | Yes |
| 6 | Is the data available? | No | Some | Yes |
| 7 | Who would approve outputs? | Nobody | Unclear | A named person |
| 8 | Can you measure it today? | No | Roughly | Yes |
| 9 | Do customers feel it? | No | Indirectly | Directly |
| 10 | Would success fund the next one? | No | Maybe | Yes |
What your score means
Pick the workflow that would pay for the next one.Intelligent AI World
Not sure where you land? The AI Readiness Map scores your workflows for you, and it is refunded if it doesn’t find one worth $200K a year for an SMB, or $500K at mid-market and up.
A worked example
| Question | A dental clinic’s missed calls | Score |
|---|---|---|
| How often? | Every day | 2 |
| Cost today? | Lost new-patient bookings | 2 |
| Cost of a mistake? | A lost patient | 1 |
| Rules-based? | Mostly: book, reschedule, answer FAQs | 2 |
| In your systems? | Practice calendar | 2 |
| Data available? | Hours, services, insurers | 2 |
| Approver? | Front-desk lead | 2 |
| Measurable today? | Missed-call count | 2 |
| Customers feel it? | Directly | 2 |
| Funds the next one? | Yes | 2 |
| Total | Build it properly | 19 |
Common first-agent mistakes
Frequently asked questions
What is the best first workflow for an AI agent?
A repeated, rules-based task with a clear cost today: missed calls, intake, scheduling, document review or claims.
How long should a first agent take?
Self-serve tools start in minutes. A built, integrated agent takes about three weeks at IAW, or 12 weeks for a full sprint.
How do I know if it worked?
Measure the task before you start, then compare the same numbers after launch.
What if my top workflow scores low on data?
Fix the data first, or pick the next workflow. A build without data is a guess.
Can IAW score my workflows for me?
Yes. The AI Readiness Map does it in two weeks, and it is refunded if it doesn’t find a workflow worth $200K a year (SMB) or $500K (mid-market and up).
Read IAW vs ChatGPT, Claude, Cowork, Manus, Kimi and Genspark before you choose.
Sources: IBM, AI trends 2026; Capgemini, Top tech trends 2026.