SF Ruby
Conf '26
The state of
agentic
engineering 2026
We surveyed engineers, tech leads, and founders at SF Ruby Conf about how they work with coding agents: which ones they use, whether they run them locally or in the cloud, and who, if anyone, reviews the code.
Who answered the survey
Just over half are engineers, and 57% work at companies of 50 people or fewer.
Role
Company size
say agents wrote most or all of the code they shipped last month
Agents write most of the code now.
54% say agents wrote roughly all of the code they shipped in the past month, and 28% say most of it. Only 1% say none of it.
How much of the code you shipped in the past month was written by coding agents?
use Claude Code every day
Nearly everyone uses Claude Code daily.
Everyone in the survey has tried it, and only 7% tried it and stopped. Codex is second, with 20% using it daily. For Cursor Agent and GitHub Copilot Agent, more respondents have stopped using them than use them now.
Which coding agents have you tried?
69% use two or more agents, daily or sometimes.
talk to their agents from a terminal
Most people still work from the terminal.
The terminal is well ahead of IDEs, which are at 41%. Agent development environments (ADEs) are much less common, at 12% local and 11% cloud.
On a typical day, what is your interface to coding agents?
34% use the terminal and nothing else.
regularly prompt agents from their phone
Fewer than a quarter regularly prompt agents from their phone.
Another 38% do it sometimes, and 38% never do.
How often do you chat with your coding agents from your phone?
pay $100 a month or more for coding agents
Most people pay $100 a month or more.
34% are on a $100–200 plan and 25% run more than one $200 plan. Together that is 59%. Another 28% are on a $20–50 plan, and 9% say their company pays. Respondents whose agents write nearly all of their code tend to be on the larger plans.
Which coding agent plans do you use?
run more than one $200 plan at the same time.
have an agent review code before a person does
Agents review first. A person still approves.
About four in five say a person must approve a change, and for 46% it has to be someone other than the person who prompted the agent. 14% merge with no human review once CI and agent reviews pass.
What is the standard code review process for features and bugfixes?
At companies over a thousand people, 83% need a second person. Among solo developers, 43% merge with no human review at all.
What is your agentic code review setup?
A quarter use several code review agents.
say non-engineers change the product with coding agents
Non-engineers are changing the product.
Most commonly (36%), non-engineers prompt coding agents and engineers review the result. 23% say they ship changes themselves, like landing pages or small fixes. About half say non-engineers use agents only for non-code work.
How do non-engineers on your team use agents?
Does your team use AI agents inside Slack or Teams?
38% use agents in Slack or Teams. At companies with more than two hundred people, it's 70%.
Teams want shared agent sessions.
Few have them.
We asked if two or more people on the team had ever worked with the same agent in the same conversation.
haven't shared an agent session, but want to
share agent sessions regularly
Excluding solo developers, 63% of people on a team want to work in shared agent sessions but don’t yet. One person's answer for how they share an agent: "Screen sharing…"
agent session looks like
Work with your team in shared agent sessions.
Shared context from ideation, through implementation, and all the way to code review and merge.



Olga is typing
Ibrahim is typingMultiplayer Agent Chat
Build together in shared agent chats. Anyone on the team can jump in.
Live App Preview
Every agent runs your app on its own cloud computer. No local checkout.
Code Review
Read the diff with QA results and agent-guided review.
How does your team compare?
The survey is still open. Answer the same questions and see where your team sits next to these numbers.
# Superconductor ran this survey at SF Ruby Conference in San Francisco, November 10–12, 2026. Answers are self-reported and come from conference attendees, a crowd that leans toward Ruby, startups, and early adoption of coding agents. Results are aggregated; no individual response is identifiable. Percentages are rounded to the nearest whole point, so some charts sum to a point over or under a hundred. "Pick all that apply" questions sum to more.
Stop
copy-pasting
your agent output.
Share agent sessions and context from ideation, through implementation, and all the way to code review and merge. Bring your Claude Code, Codex, or other subscriptions.