Everyone's shipping AI-written code now. The edge is proving it works
A survey out this month puts AI coding-agent use at 39% of developers. That number should change how you pick a software shop. The risk isn't that your vendor uses agents. It's that you can't tell the ones who verify from the ones who don't.
I read a survey this month that should change how you buy software. JetBrains puts AI coding-agent use at 39% of developers. A year ago that number was a curiosity. Now it's the floor. Any boutique or internal team building software for you is using agents to scaffold, refactor, and write tests, whether they put it in the proposal or not.
So the old buying question, "do they use AI?", is dead. The new one is sharper: how do you know the agent's code is correct?
The trap most buyers walk into
Here's what bugs me about how this is landing. Teams ship agent-generated code faster, and the verification step is the first thing that gets dropped under deadline pressure. An agent writes a function, it passes the happy path, everyone moves on. The edge cases, the security review, the "did this quietly break the feature from last week" check all get skipped. That's where unverified agent code rots.
The research backs the worry. LTM's SDLC AI Radar calls governance and the ongoing evolution of AI-built software a requirement now, not a nice-to-have. The smart money is on "someone has to check the agent," not "the agent is fine on its own."
What we actually do differently
This is the part I'd want a buyer to hear. At iotasol we formalized an agentic-SDLC practice on purpose. Agents do the boilerplate and the boring first draft. Humans review. And, this is the bit most shops skip, we run an evaluation harness on the agent's output: tests that catch regressions, a checkpoint before anything merges, a record of what changed and why.
It isn't slower. It's how we keep a lean Melbourne team competitive against shops charging enterprise rates. Local engineers are running a $105K to $145K median. Agents let us deliver more per head without padding your invoice with bodies we don't need. You get the speed and the senior eyes.
Why this is local, and why now
Two Melbourne-specific reasons this matters this quarter. First, the City of Melbourne runs business-growth-impact grants that fund exactly this kind of agentic-SDLC R&D and AI-governance IP. We're a local firm that can put that runway to work building tooling you'd benefit from. Second, more mid-market Australian firms are commissioning bespoke software as the enterprise market surges past $1.44T, and the ones who get burned are the ones who never asked their vendor the verification question.
The through-line
AI coding agents aren't a differentiator. Verifying their output is. Next time you brief a software team, skip "do you use AI" and ask "show me how you prove the code is sound." If they hesitate, that's your answer.
We built our agentic-SDLC practice so we never have to hesitate. If you're commissioning bespoke software in Melbourne and want a team that ships fast and shows its working, let's talk. Book 30 minutes and bring one project you're nervous about.
You Dream, We Build.