Jira went AI-native. Developers asked for the opposite.

On July 15, Atlassian announced it was evolving Jira "for AI-native software development," a system where every agent action is "visible, governed, and tied to a business outcome." The pitch is that agents now do a lot of the work, so the tracker should plan, assign, govern, and measure that work across humans and machines alike.

Then the developers showed up in the comments.

The top reactions were not gratitude. They were tired. One went, roughly, "so now not only do I deal with the eternal junior, I now have to document everything I get it to do." Another, blunter: people who used to love Atlassian's tools calling the new direction patronizing. A third pointed at the velocity math directly: AI usage up sharply, developer velocity barely moved, and the proposed answer is more AI. That gap, more process sold as less, is the whole reaction in one line.

This post is not a dunk on Jira. Atlassian ships real engineering, and "govern the agents" is a genuine enterprise need for teams that have to answer to auditors. The point is narrower and it is about who the tool is for. When the people who write the code react to an AI-native tracker with fatigue instead of relief, that is a signal, and it is the signal Radial is built around.

#What developers actually asked for

Read enough of these threads and the ask is remarkably consistent. It is not "give me a smarter ticket." It is closer to the opposite.

• A tool that stays a tool. A QA director who administers Jira for a game studio put the lock-in feeling exactly: it gives you "just enough to almost get it how you want it but never quite enough to really nail it," with the simple things "locked away behind either marketplace addons or just never implemented." That same person called AI-powered tracker features the thing "everyone is pitching and no one wants."
• The agent, not the app, holds the intelligence. On a thread about Jira going AI-native, one developer's summary of Rovo was that it is terrible, and that the thing that actually works is handing an API key to their own agent, which "gets you exactly what you need." That is the tell. The intelligence people want is the agent they already run, pointed at a tracker it can read and write. Not a copilot bolted into a sidebar the vendor owns and meters.
• Honesty over cleverness. In a separate debate about what an agent should do when it cannot reach the issue tracker, the line that got the most agreement was about "the honesty of the run." Devs would rather the agent say "I could not read the issue, here is what I need" than silently invent a data shape that almost matches. The almost is where the bugs hide.

None of that is anti-AI. Every one of those developers runs agents daily. What they are rejecting is a specific move: turning the tracker itself into an AI product, with a copilot to learn each quarter and a meter to watch drain, when all they wanted was a fast place to file the work their own agent is doing.

#The distinction that matters: AI in the workflow, not in the tool

There is a real line between two things that get blurred on purpose in launch copy.

One is AI in your workflow. Your agent, your model, your keys, driving a tracker through real interfaces. This is good, and Radial is built for it. Agents get a first-class credential and ride free, because a credential is not a seat.

The other is AI in the tool. A copilot the vendor ships into the product, priced as its own SKU, metered by usage, updated on the vendor's schedule so the buttons move every quarter. This is the thing developers keep saying they did not ask for.

Radial takes the first and refuses the second. The tracker stays a fast, trustworthy system of record. The intelligence belongs to the agent you already trust, reaching the tracker over the CLI, the REST API, or MCP. That way an agent gets exactly the "hand it an API key and it does what you need" experience developers described wanting, without a metered copilot in the middle.

Here is what "hand your agent a scoped key" looks like in practice. This creates a triage-only key an agent can use to file and sort issues, and nothing else:

# create one scoped key per agent (read-only, triage-only, or full)
radial keys create "triage-bot" --scope triage --json

# the agent files an issue with that key, from anywhere it runs
RADIAL_API_KEY=rk_... radial create "Investigate 500s on checkout" \
  -t ENG -p high --json

One key per agent, revocable in one click, scoped to exactly what that agent should touch. The agent does the thinking. The tracker just holds the record, fast.

#The pledge that says it out loud

Most tools will tell you AI features are optional, then ship them into the default view and add a meter next to them. We would rather write the promise down where you can hold us to it.

Radial's Plain Software Pledge: the day we ship a copilot, meter your usage, or charge you for AI you did not ask for, your subscription is free.

That is not a marketing line. It is a falsifiable, dated guarantee, and it is why the pricing is what it is. Radial is $50 per user, per year, flat, billed annually. Agents ride free, because every agent credential is excluded from billed seats by design. No trial, no free tier, no credits to pre-buy, no usage line on the invoice. The rate you join at is the rate you keep. When your throughput doubles because your agents got faster, your bill does not move.

The counter-signal is worth naming honestly: the trade press keeps framing AI-in-tooling as inevitable infrastructure, baked in automatically whether you asked or not. Maybe that is right for some products. It is not right for the one place your work is recorded. A system of record should be predictable, fast, and boring on purpose. That is a choice, and the pledge is us making it in writing. The meter that arrives in an update is the same move in a different costume, which is a story worth reading on its own.

#FAQ

#What did Atlassian announce for Jira and AI?

On July 15, 2026, Atlassian announced new Jira and Teamwork Graph capabilities for "AI-native software development," positioning Jira as the place to plan, assign, govern, and measure work across both humans and AI agents, with agent actions made visible and tied to business outcomes. It builds on Rovo, Jira's in-product AI assistant.

#Is Radial anti-AI?

No. Radial is bring-your-own-agent, not anti-AI. The difference is where the intelligence lives. Radial has no copilot, no AI billed as a SKU, and no metered usage inside the product. Your own agent (Claude Code, Codex, or anything that speaks HTTP) drives the tracker through the CLI, REST API, or MCP, and agents ride free.

#Can my coding agent create and manage issues in Radial?

Yes. Radial exposes a 30-command CLI, a REST API, and a remote MCP server with 41 tools, all authorized by scoped, per-agent keys. An agent can file, triage, groom, comment on, and close issues, or plan a whole cycle, without a human in the loop. You give each agent a key scoped to read-only, triage-only, or full access, and revoke any one of them in a click.

#How much does Radial cost, and does AI cost extra?

$50 per user, per year, flat, billed annually. There is no separate AI tier, no credits to buy, and no usage meter. Agents do not count as seats, so they are free. The Plain Software Pledge makes it binding: if Radial ever ships a copilot, meters your usage, or charges for AI you did not ask for, your subscription becomes free.

#What if I want to leave later?

Your data leaves in one command, JSON or CSV, always available. Easy export is the point, not a loophole. A tool you cannot leave is a cage no matter how calm it looks, so the export and the pledge are what make "the tracker stays a tracker" trustworthy.

If you are re-evaluating your tracker because the AI-native pitch felt like more process instead of less, that is the reaction Radial was built for. Read the manifesto for the full stance, or wire your own agent into the tracker over the developer surface. The tracker stays a tracker. Your agent does the thinking.