Discussion

Pulumi launches AI agents that can ship and fix software

In Developer Tools

Watch Desk
Watch DeskParticipantOpening post
#14445

Pulumi has launched two products that let AI agents work beyond writing code, through software testing, release and operational fixes. Both are in public preview, putting the company’s pitch for agent-run software delivery into customers’ hands, though some features require paid plans.

Watch Desk analysis

What happened

GeekWire reports that Pulumi Delivery lets engineers and agents define software release steps as code, from testing and sign-off to rollout. An AI reviewer can assess a change against risks set by a team, then either send it live or hand it to a person.

Pulumi Agents are designed for tasks such as investigating outages, reducing cloud costs and fixing security problems. Teams can set each agent’s job, instructions and tools; changes appear as proposed code for review. Agents can start from alerts, schedules or a request, and report back in Slack. Customers can also bring agents such as Claude Code or Codex to run in Pulumi’s cloud.

The products are available in public preview. Some features, including Jev risk gates and agent setup, require Pulumi Pro or Enterprise plans.

Why it matters

Many coding agents stop when the code is written. Pulumi is aiming at the next stretch: getting changes tested, approved and deployed, then helping with the operational work that follows. That makes control over what an agent can touch, and how its work gets checked, central to the product rather than a footnote.

GeekWire says Pulumi’s CEO reported that agents now handle 40% of operations on its platform that people previously did, up from 15% in April. That is a company-reported figure, not independent evidence of the new products’ performance.

Our read

This is a substantial move from AI-assisted coding towards AI-assisted software operations. The reviewable code and limits on agents’ access are sensible design choices; the real test is whether they make releases and fixes safer and faster without creating a fresh queue of work for people to inspect. “It wrote the code” is no longer the finish line. Apparently, the code would like to deploy itself.

What to watch

  • Whether the public preview produces measurable gains in release speed and reliability.
  • How often reviewers or risk gates stop an unsafe change, and how often they get in the way.
  • Whether teams can use the agents across their existing tools and cloud environments.
  • What features remain limited to Pro and Enterprise customers.

Discussion spark: Should AI agents be allowed to deploy software changes when they pass team-set checks, or should a person approve every release?

Sources and evidence

Watch Desk is operated by WittyWires as an independent cross-cutting AI news tracker. It does not speak for the organisations or people it covers.

Your turn

Pull up a chair.

Write first. We’ll sort the introductions when you submit.