A lot of business owners are asking whether they can skip the developer and have AI build their software. For a simple internal tool, sometimes you can. For a system your business depends on every day, the answer is usually not on your own yet. The best evidence I’ve seen for that comes from Anthropic’s June 2026 study of how people actually use Claude Code, and it matches what I see in my own work.
What the study looked at
Anthropic analyzed about 400,000 interactive Claude Code sessions from roughly 235,000 people between October 2025 and April 2026 (Anthropic). Each session was rated for the user’s expertise in that specific task, based on how precisely the user gave direction, what they asked the AI to verify, and whether the user corrected the AI or the other way around.
What they found
The people still made most of the important calls. Users made about 70 percent of planning decisions, the “what to do,” while Claude made about 80 percent of execution decisions, the “how to do it.”
Expertise made a big difference in outcomes. Verified success, meaning the session succeeded and there was hard evidence such as a commit, passing tests, or the user confirming the result, rose from 15 percent for novices to 28 to 33 percent for intermediate and expert users. Novices abandoned 19 percent of sessions, compared with 5 to 7 percent for everyone else.
One finding surprised me. People in software-related occupations had only a small edge over people in other professions, about 30 percent verified success compared with 26 percent. Anthropic’s broader point was that expertise in the domain, not just coding skill, amplifies how well the tool works.
Why this matches my experience
AI is very good at execution. Give it clear direction and it will produce a lot of working code quickly. The hard part is knowing what to ask for, spotting when something looks right but isn’t, and understanding the business well enough to know what matters.
Michael and I bring a combined 20 years of leading engineering teams and advising businesses on software (About Hello Spark). Some of my favorite early work was on a US tax calculation engine, where small mistakes could have big consequences. That kind of experience doesn’t go away when AI writes more of the code. It becomes what decides whether the result is any good.
What the study doesn’t say
I’d be careful not to stretch the findings. The study measured whether coding sessions succeeded. It didn’t measure whether the resulting software held up in production for months. Those are different questions, and the second one is the one that matters to a business.
AI app builder or developer?
AI app builders are a reasonable choice for a simple form, a personal tracker, or a quick internal dashboard where a mistake won’t cost much. I’d bring in an experienced developer when the software handles money, customer data, or health information, when it connects to other systems, or when your team will rely on it every day. In those cases, the hard part isn’t generating code. It’s knowing whether the code is right and what happens when something fails.
What this means for businesses
Someday you may be able to tell an AI, “Build an inventory system for my business,” and get exactly what you need. We aren’t there yet.
For now, the best results come from people who understand both the business problem and how good software gets built. In the right hands, AI makes it possible to build custom systems that compete with best-in-class products, for a fraction of what that used to cost.
If you’re evaluating someone to build AI software for you, ask about their experience with the kind of problem you have. That will tell you more than which AI tools they use. These 10 questions to ask an AI development agency are a good place to start.