Live survey · open now
State of AI agent coding 2026
How developers really work with AI agents: which tools, which models, how many agents at once, how much code they still type, and what job interviews ask now. Answer about 25 questions in three minutes and get your type.
No email, no account. Kin asks, you tap, the card is yours.
What it asks
About 25 questions, five sections, three minutes
- 1
You
Role, years paid to write software, team and company size.
- 2
Your tools
Which tools you use weekly, your main one, model families, the model for the hardest problems, reasoning effort, who pays.
- 3
Working with agents
Agents at once, how you run several and the hardest part of it, how often one works unattended, AI in code review, the share of code AI wrote, when you last typed a line.
- 4
What changed
New kinds of work since AI, how you feel compared with before, languages.
- 5
The job market
Your company’s stance, what interviews asked, what hirers ask, job security against two years ago.
The last two: your region (optional) and one line of advice for a developer who has not started with agents, which can appear on this page if you allow it. Where a list misses your tool or setup, pick “Something else” and type it in.
The result
Five types, from how you work with agents
The Conductor
Four or more agents, hands off, eyes on the board.
You stopped writing most of the code and started running the people who do, except they are agents. Your day is prompts, reviews and deciding what runs next.
The Delegator
Agents work while you do something else. You check in, not out.
You hand a task over and walk away, then come back for the result. Not everything runs at once, but what runs, runs without you watching.
The Juggler
Several agents at once, and you watch every one of them.
Two, three, six terminals, all moving, all with your eyes on them. You get the speed of parallel agents and pay for it with attention.
The Pair Programmer
One agent, one task, your eyes on every line.
AI sits next to you rather than instead of you. You read what it writes, keep the taste in your own hands and ship with it, not through it.
The Holdout
AI stays in the chat window, or out of the building.
By choice or by rule, agents are not writing your code. You might ask a chat a question now and then, but the keyboard is yours.
Plus the Generalist badge for anyone doing three or more new kinds of work since AI, and three numbers that place you against everyone who answered.
The data
The results open soon
The charts, the filters and the CSV appear here once enough developers have answered that no single answer stands out.
Be in the first wave.
Answer now and your answers are in the first charts. Your type and your own numbers come right away, on your result.
Add your answersHow it is run
Honest numbers, no tracking
Who can answer
Anyone, once per browser; a bot check runs when a quiz is finished. There is no account and no email step. The sample is developers who found this page, so it leans toward people who already use agents; the filters let you read it for your own role and experience.
What is kept
The answers and the result under a random id, the day, and phone or desktop. No IP address, no location from the network, no cookies. The browser remembers the run in its own storage so a closed tab loses nothing. The privacy page has the full list.
What is published
Counts only. The charts open from at least 10 answers, the hiring chart from 20 hirers, a filtered view needs 30 people, and in a filtered view options under 5 people show as 0%. The counts will also be published as a CSV you can chart yourself.
Questions
About the survey
Who is this survey for?
Developers who use AI at work in any form, from autocomplete to a fleet of agents, and the managers and founders who hire them. People who use no AI tools are welcome too; the quiz is shorter for them and their answers matter for the comparison.
What do I get for three minutes?
Your type (Conductor, Delegator, Juggler, Pair programmer or Holdout), three numbers that place you against everyone who answered, a read on what hirers ask about agents now, and a card you can share. Then the live report, with every chart filterable by role, experience and main tool.
What is stored about me?
Your answers and the result, under a random id, with the day and whether it was taken on a phone. No name, no email, no account, no IP address and no location; region is a question you can skip. An email address is kept only if you type one to get the report, apart from the answers.
How are the charts kept honest?
The charts open once at least 10 developers have answered, the hiring chart once 20 people who hired this year have, and a filtered view needs 30 people in it. In a filtered view any option picked by fewer than 5 people shows as 0%. Questions shown only to some people say so. The counts are published as CSV.
Why does a dashboard for AI agents run a survey?
Briefkin shows what AI agents are doing on one board, and the people who need that run several agents at once. This survey measures how many developers work that way, what the rest use instead, and what the job market asks of them. The results are public for everyone to use.
Can I use the numbers?
Yes. Quote them with a link to this page (each chart has a "Copy this stat" button with the sentence and its anchor) or download the CSV. The survey is by Briefkin; the data is yours to chart.
Running agents in parallel? See them on one board.
Briefkin puts every Claude Code, Codex and Cursor session on one board: working, waiting on you, done. Answer them from your phone, and see what each one shipped.
Free during the open beta. No card.