Anshad Ameenza.
Productivity··Updated: Nov 21, 2026

The Careful Person's First Month With AI Agents

A week-by-week plan for careful experts adopting AI agents: which tools to try first, what to measure, and how fast adopters can build the judgment AI lacks.


Every team has two people who worry about AI for opposite reasons. One has used it every day since 2023 and ships a lot. The other hasn’t touched it, and is the person everyone asks when the numbers look wrong.

Which of them is better placed for the next five years? We don’t know yet, and the research is thinner than the confidence on either side. What we can say is that the two of them have different next moves, and that one of those moves is much easier than the other.

This post argues that the careful expert who avoids AI has the cheapest, highest-return move available to anyone right now: adopt, and bring the checks along. The fast adopter’s move is harder, because no model hands out judgment. Below is a four-week plan for the first person, and a practice routine for the second.

One controller, before her first month

Ruth is a financial controller at a mid-sized distributor. She is a composite, built to show the shape of the situation, not a real person. She has run month-end close for years and has never used an AI tool for work. Her team trusts her because she catches things.

Ruth’s month-end close tasks and the check she runs on eachFour stacked cards. Variance commentary, checked by tracing every number to the ledger. Bank reconciliations, checked by tying to the statement to the cent. Accrual estimates, checked by a second reader who signs off. Board memo, checked by reading it aloud before sending. A footer says AI hours this year are zero, and every check is already in place.Ruth’s close, and the check on each taskvariance commentarycheck · every number traced to the ledgerbank reconciliationscheck · ties to the statement, to the centaccrual estimatescheck · a second reader signs offboard memocheck · read aloud before it is sentAI hours this year · 0checks already in place · all four
Illustrative: Ruth already runs the review layer that AI output needs. The only thing missing is the generator.

Look at what she already has. Every task ends in a check, and every check is something a person does to output they didn’t fully trust. That is exactly the work AI output demands. Ruth has spent her career building the half of the system that most fast adopters are missing.

What she lacks is the cheap first draft. Her variance commentary takes most of a day because she writes all forty lines herself.

The model: judgment and AI use are separate axes

Here is the model in one sentence: whether you use AI and whether you have good judgment are independent, and AI multiplies whatever judgment is already there rather than supplying any.

A two by two grid of judgment against AI use, with two moves toward high leverageColumns are no AI and uses AI. Rows are strong judgment and weak judgment. Strong judgment with no AI is reliable but slow. Strong judgment with AI is large leverage. Weak judgment with no AI is low output. Weak judgment with AI is confident errors at scale. An amber arrow runs along the top row from left to right: move one, adopt and keep your checks, takes weeks. A cobalt arrow on the right runs from bottom right up to top right: move two, build judgment, takes months.Two axes, not oneno AIuses AIstrongjudgmentweakjudgmentreliablebut slowlargeleveragelowoutputconfident errorsat scaleamber · move 1 · adopt, keep your checks · weekscobalt · move 2 · build judgment · monthsAI multiplies the judgment that is already there
The two moves into the top right are not symmetric. Adopting a tool takes weeks. Building judgment takes months of deliberate review.

The careful person’s suspicion is the common position, not an eccentric one. The 2025 Stack Overflow Developer Survey found 84 percent of developers using or planning to use AI tools, while 46 percent said they distrust the accuracy of the output and 3 percent said they highly trust it. Almost everyone is using tools almost nobody trusts. The person who arrives with working checks is the one equipped for that situation.

Move 1 is a tooling problem. Move 2 is a skill problem. Tooling problems are cheaper.

Week 1: a chat assistant on drafts nobody audits

The week one plan: tools, tasks, the carried-over check, and the start of a logTop card lists tools: ChatGPT, Claude, Gemini. Second card lists tasks: a recap from your own notes, a first draft of a routine email, a policy paragraph rewritten plainly. Third card, the check: only drafts you would reread line by line anyway. Bottom is a log with four columns: task, minutes alone, minutes with AI, errors caught. One illustrative row: recap, 25, 10, 2.Week 1 · chat assistant · low stakestoolsChatGPT · Claude · Gemini, any one of themtasksa meeting recap from your own notesthe first draft of a routine emaila policy paragraph, rewritten plainlythe check you bringonly drafts you would reread line by line anywaytaskalonewith AIerrors caughtrecap25 min10 min2the log is the point · one row per taskillustrative row, not a measured result
Week 1 is about calibration, not savings. Use drafts you would reread anyway, and start the log on day one.

Pick one chat assistant and stay with it all week. Switching between three is a way to feel busy without learning any of them. The tasks should be ones where a mistake costs you a minute of embarrassment at most, and where you would reread the whole thing regardless.

The log matters more than the drafts. Write down how long the task usually takes, how long it took with the assistant including your edit, and every error you caught: a wrong date, a softened commitment, a summary that dropped the one point that mattered. Ruth’s first recap invented an action owner who wasn’t in the meeting. She would have caught that anyway. Now she has it on paper, and that is how you learn what this particular tool gets wrong.

Week 2: a task you could do yourself, done both ways

Results from the 2023 BCG field experiment, and the week two tool choicesFrom Dell’Acqua and colleagues, a 2023 Harvard Business School working paper with 758 BCG consultants. On tasks inside the AI frontier, consultants with AI completed 12.2 percent more tasks. On a task designed to sit outside it, consultants with AI were 19 percentage points less likely to reach the correct answer. Below, week two tools: for code, Claude Code, Cursor or GitHub Copilot. For data, the assistant in Excel or Google Sheets. The rule: do the task yourself too, then compare.758 BCG consultants, with and without AIinside the frontier · 12.2% more tasks doneoutside it · 19 points less likely correctDell’Acqua et al. · 2023 · HBS working paperif you write codeClaude Code · Cursor · GitHub Copilotif you live in spreadsheetsthe assistant inside Excel or Google Sheetsdo it yourself too · then compare the twolog time both ways and every difference you find
The frontier is jagged. Inside it, AI made consultants more productive. Outside it, AI made them more often wrong, and the tasks looked alike.

In 2023, Fabrizio Dell’Acqua and colleagues ran a randomized experiment with 758 consultants at Boston Consulting Group. On creative product tasks that GPT-4 handled well, consultants with AI finished 12.2 percent more tasks, faster and at higher rated quality. On a business problem built to sit just outside what the model could do, the AI group was 19 percentage points less likely to get the right answer. The authors called the boundary a jagged frontier, because tasks of similar difficulty landed on different sides of it.

You can’t see the frontier from the outside. You find it by doing a task you know cold, once by hand and once with the tool, and diffing the results. A developer might hand Claude Code or Cursor a small refactor they would normally do in an afternoon. Ruth asks the spreadsheet assistant to build the accrual roll-forward she has built for years, then ties it to her own version.

The week’s measure is a second column of errors: the ones in the AI version that your own version didn’t have. That column is your personal map of the frontier, for your work, with this tool.

Week 3: deep research, with every citation opened

A source-checking flow for a deep research reportA vertical flow. A deep research tool such as ChatGPT deep research or Gemini Deep Research returns a report with citations. Step one, list each factual claim. Step two, open its citation. Step three, ask whether the source says it. Three outcomes: says it, says something weaker, or does not say it. The week’s measure is the share of claims the sources support.Week 3 · check the sources, not the prosea report with citationsChatGPT deep research · Gemini Deep Research1 · list each factual claimnumbers, dates, names, quotes2 · open the cited sourcethe page itself, not the snippet3 · does it say that?mark one of three outcomessays itsays lessdoesn’t say itthe measure · share of claims a source supports“says less” is the miss careful people catch
A cited report is a list of claims to check, not a finished answer. Count how often the source says what the report says it does.

Deep research tools, such as ChatGPT’s deep research mode and Gemini Deep Research, search the web for a while and return a long report with citations. The citations are what make them useful to a careful person, because they turn the report into a checklist.

Pick a question you already know something about. Ruth asked for the current treatment of a lease accounting edge case she had dealt with the year before. Then open every citation and mark it. The middle outcome is the one to watch for: the source exists and is relevant, but it says something weaker or narrower than the report claims. A fast reader skims past those. A careful one has been catching that exact move in other people’s memos for years.

Week 4: delegate a multi-step task, with gates

A multi-step delegated task with approval gates between stepsFour agent steps alternate with four human gates. The agent writes a plan, and you approve the plan. It pulls last quarter’s workbook and drafts the reconciliation, and you review the changes. It flags variances over a threshold, and you check two of them by hand. It writes the summary, and you sign before anything is sent. A footer notes that Claude Code, by default, asks before editing files or running commands.Week 4 · agent steps, human gatesagent · writes a plan for the whole taskgate · you approve or edit the planagent · pulls the workbook, drafts the recongate · you review every change it madeagent · flags variances over the thresholdgate · you redo two of them by handagent · writes the summarygate · you sign before anything is sentClaude Code asks before edits and commands by default
Delegation is safe when every irreversible step waits for a person. Approve the plan, review each diff, and never skip the last gate.

By week four you have a log that says where your tool is reliable and where it isn’t. Now hand it a task with several steps and put yourself between them. Agentic tools already support this shape. Claude Code, by default, asks for confirmation before it edits a file or runs a shell command. Coding agents in Cursor and GitHub Copilot propose changes you review before accepting. Outside code, the same pattern works with a document or spreadsheet agent if you split the job and refuse to let it run end to end.

The measure this week is where you stopped it. Count the gates where you rejected or rewrote the agent’s work, and note how much rework that saved. Ruth stopped hers at the variance gate: it had flagged a timing difference as a real variance. The two she redid by hand are what told her to look.

The harder move: building judgment after adopting fast

Qualitative results from the 2026 persistence trials, and a four step review repFrom randomized trials with 1,222 participants by Liu, Christian and colleagues, reported by Berkeley News in October 2026. Relative bars with no exact numbers: people without AI as the reference; people who asked for hints stayed close to them; people who asked for direct answers performed worse and gave up sooner once the AI was removed. Below, a four step review rep on a line: predict the answer yourself, generate with the model, diff the two, and log every gap.After the AI was taken awaynever had AI · the referenceasked for hints · close to the referenceasked for answers · gave up soonerLiu, Christian et al. · 2026 · 1,222 people · relative onlyThe review rep, once a daypredictyours firstgeneratethe model’sdiffwhere they differlogwho was rightthe log becomes your own error map
How you use the model decides what you keep. Hint-seekers held up when the AI was removed; answer-seekers gave up sooner.

The fast adopter’s problem shows up in the best-known field study of AI at work. Erik Brynjolfsson, Danielle Li and Lindsey Raymond studied a chat assistant rolled out to more than five thousand customer support agents (NBER working paper in 2023, published in the Quarterly Journal of Economics in 2025). The newest and least skilled agents gained the most, while the most experienced saw small speed gains and slight quality declines. The authors’ reading is that the tool spread the practices of the best agents to everyone else. On scripted work, a novice can borrow judgment from the tool. That is real, and it is also borrowed. It covers the cases the top agents already solved and nothing past them.

So the fix can’t be more tool. In October, Berkeley News reported on randomized trials by Grace Liu, Brian Christian and colleagues with 1,222 participants solving math and reading problems. When the AI was removed, people who had mostly asked it for direct answers did worse and gave up sooner than people who never had it. People who asked for hints stayed close to the control group.

The practical version is the review rep. Before you prompt, write your own answer, even a rough one. Generate. Diff the two. Log who was right and why. Ask for hints rather than answers on anything you want to keep being able to do. It is slower, deliberately, for a few months, and it gives a fast adopter what a careful expert built over years: a record of where they are wrong. For the engineering version of this, where review becomes most of the job, see the engineer after the editor.

What this gets wrong

Three limits of the two-axis model and the four-week planThree stacked cards. Judgment is per domain: strong on the ledger can be weak on the code. The evidence is narrow: one firm and a 2023 model in the frontier study, short sessions in the persistence trials. You grade yourself: the log is only as honest as the person keeping it.judgment is per domainstrong on the ledger can be weak on the codethe evidence is narrowone firm, a 2023 model, short lab sessionsyou grade yourselfthe log is only as honest as its keepera direction to test, not a measured law
The model is useful and too clean. Judgment is uneven, the evidence is narrow, and you grade your own log.

Judgment isn’t one number. Ruth’s checks are excellent inside accounting and worth little if she asks an agent to write a script against the general ledger’s database. A careful expert outside their domain is a fast adopter with good manners. The four weeks should stay inside the work you already know how to check, and the plan says nothing about the rest.

The evidence is narrower than the model. The frontier study used one firm and a 2023 model; the frontier has moved since, and probably not uniformly. The persistence trials measured people right after a short session on math and reading, which says little about a year of daily use. The support study covered scripted work, where the best answers already existed to be copied.

And the plan relies on a log you keep yourself. A careful person can still go easy on a tool they have started to like. If you can, have a colleague spot-check two entries a week. Then there is the chance that the axes stop being independent: tools may get good enough at checking their own work that the careful person’s edge shrinks. If that happens, the annual review in the annual review for an AI year is where you would see it first.

Your next move

If you are the careful one, block thirty minutes on Monday, open one chat assistant, and hand it the most boring draft on your list. Start the log before you start the draft. Four weeks from now you will have something the fast adopters around you mostly don’t: a written record of where the tool is right, where it is wrong, and how you know.

If you are the fast one, write your own answer before your next prompt. Do it once a day for a month, and keep the log. For why the shape of the help matters, ten minutes of answers goes deeper on the persistence trials.

AI AgentsAI AdoptionJudgmentProductivityReview
Share:
Anshad Ameenza
About the Author

Anshad Ameenza

Lifelong Learner, Engineer, Technology Leader & Innovation Architect

20+ years of experience in technology leadership, innovation, and digital transformation. Building and scaling technology ventures.

Only if you find it useful

No pitch here. If these pieces are worth your time, you can get new ones in your inbox. If not, skip it with a clear conscience, nothing is being sold. Rare emails, no spam, leave whenever you like.

Continue Reading

Related Articles