Software Development Deflation Is Real

Their answer: “No.”

“Why not?”

"Good code costs money."

He's not wrong. Good code does cost money. But he's making an argument about quality to defend a price point. And the price point is what's collapsing.

Boris Cherny runs Claude Code at Anthropic. He hasn't written a line of code by hand in over two months. He ships 22 to 27 completed code updates a day. All AI-generated. All reviewed and approved.

An OpenAI researcher was asked what percentage of his coding is done by AI. His answer: "100%. I don't write code anymore."

Across Anthropic as a company, 70 to 90 percent of all code is AI-generated. Claude Code itself was 90% written by Claude Code.

These are the best software engineers on the planet. They're building the most sophisticated AI models in history. And they decided that writing code by hand was slowing them down.

We take SMB owners from using ChatGPT to replace Google searches to building agentic systems in 4 weeks. We are kicking off our third AI for Business Owners Cohort April 13th.

We’re offering 28% off normal prices if you sign up before March 15th.

Here's the proof: model release cadence. In late 2025, Anthropic shipped three major model updates in three months. In early 2026, two more in two months. The pace is accelerating. The models are getting better. They're doing it with the same or smaller teams.

If not writing code produced worse results, you'd see the opposite. Slower releases. Buggier models. Instead, you see the fastest, highest-quality output cycle in the history of software.

So when a developer tells a small business owner that "good code costs money" and uses that to justify a $300K project, he's right about one thing: quality matters. He's wrong about what quality costs. The people who literally invented these coding tools proved that agentic development is faster and produces better results. That developer is defending his margin, not his quality standard.

I run AI cohorts for business owners. Last week, a participant described his company's recruiting pipeline. Specialized staffing. They source candidates, look up contact info, de-dupe against their applicant tracking system, upload to HubSpot, and run outreach campaigns. Three people do this full time.

That morning, I sat down with Claude Code and built the full automation proposal. API connections to the industry directories. Phone and email lookup services. De-duplication against their ATS. Bulk upload to HubSpot. I packaged it in a branded pitch deck.

It took about twenty minutes. A year ago, scoping that project alone would have been a $10-15K engagement from a consulting firm. Building it would have been another $200K to $300K. Today, the scoping is essentially free if you know how to ask. And the build is compressing fast. I'm seeing projects that were $300K quotes 18 months ago coming in at $30K now. That number is still falling.

This isn't because the work got simpler. It's because the labor structure changed. Discovery that took weeks takes hours. Architecture decisions that required senior engineers are handled by someone who knows the right questions to ask plus Claude Code. The coding itself, line by line, is the part AI is best at.

You still need someone who understands your business problem. You still need code review. You still need good architecture. But "good code costs money" doesn't mean what it meant two years ago. The cost of good code is dropping every month.

If you're a business owner sitting on a $300K software quote, pause. Not because the project isn't worth doing. But because the cost to build it is deflating in real time.

Here's the uncomfortable truth: most developers haven't come to terms with how fast AI coding has improved. They're like everyone else. They tried it six months ago, got spaghetti code, and wrote it off. Developers are technical, but that doesn't make them immune to getting stuck in their ways. It is genuinely hard right now to find people building in the new way.

That's the gap. The tools have moved faster than the people who use them.

Which is why we're focused on two things at OwnerRx. First, training business owners on AI so they become better consumers of technical work. When you understand what these tools can do, you stop accepting $300K quotes without pushing back. You ask better questions. You spot when someone is defending a margin instead of delivering value.

Second, training internal teams to build the systems they would have hired outside developers for eight months ago. Your ops person who knows your business inside out is often a better candidate to scope and direct an AI coding tool than an outside developer who doesn't know your workflow.

Three things to do right now:

1. Pressure-test your quotes. Ask your developer if they're using AI coding tools. If they're not, ask why. If the answer is some version of "good code costs money," push harder. The best engineers in the world disagree.

2. Scope it yourself first. Spend twenty minutes with Claude or ChatGPT describing your workflow. You won't get a finished product, but you'll know 80% of what's possible before you write a check.

3. Invest in your own people. The developer shortage everyone talks about looks different when your existing team can direct AI tools. The bottleneck isn't coding talent. It's people who understand your business and can talk to the machine.

Anthropic's CEO said we're six to twelve months from AI handling most software engineering end to end. I think that's aggressive for complex enterprise systems. For the kind of custom software most small businesses need? We're already there. The gap is knowing how to direct it.

That's what we're building at OwnerRx. Our next Builders Course starts in April. If you want to learn how to scope, build, and review systems with these tools instead of writing $300K checks, reply and I'll send you the details.

Alan

PS We are taking on some projects working with internal teams to train them in agentic coding by building something the business needs as they learn over 4-8 weeks. Book some time with me if you are interested.