AI Can't Save Software Engineering

Lately, many people have been talking about AI for Software Engineering (AI4SE), hoping to find the next big startup opportunity there.

My view is straightforward: this path leads nowhere.

Most of these ventures make a fundamental mistake — they try to use a technical hammer to hit a management nail.

That’s doomed from the start.

1. The Core of Software Engineering: Managing People, Not Code

We like to think software engineering is about code, architecture, and tools.

Wrong. At its heart, it’s a socio-technical system — a web of people, incentives, and responsibilities that ultimately shapes the software.

Conway’s Law already said it best: your software architecture mirrors your organization’s structure.

If your company is siloed, with conflicting KPIs between teams, no AI-recommended microservice architecture will save you — implementation will be chaos.

AI can highlight technical debt, but it can’t decide who should pay it down.

AI can optimize your code, but it can’t fix a team that refuses to refactor because their only KPI is “shipping faster.”

Most software engineering problems originate not in the codebase, but in the meeting room and the KPI spreadsheet.

No matter how powerful AI becomes, it can’t rewrite your company’s power structure or incentive system.

2. The Value Black Hole: You Can’t Prove It’s “Useful”

That’s the Achilles’ heel of every so-called productivity tool — AI included.

Say your AI tool detects a potential bug. How do you measure its value?

If you don’t fix it — zero value.

If you do fix it — how do you prove it would have caused a production outage? And how much would that have cost — $10,000 or $1 million?

The causal chain is too long and too fuzzy, so it all collapses into faith.

And the people paying the bills don’t care about “code coverage” or “cyclomatic complexity.”

They care about revenue, profit, and market share.

Tell them you’ve “improved engineering culture,” and they’ll ask, “So — did our R&D costs go down?”

Most AI4SE tools end up selling supplements, not painkillers — nice-to-haves, not must-haves.

And in a downturn where every company is cutting costs, you can imagine how well the supplement business goes.

3. The Business Model Mismatch: “Experts Selling to Experts” Can’t Scale

Profitable businesses usually share one trait: insiders sell to outsiders, exploiting an information gap.

Salesforce sells complex CRM systems to sales teams. Shopify sells full-stack e-commerce to small business owners.

But AI4SE? It’s experts selling to experts.

Your customers — engineers — often know as much as you do, maybe more.

Try telling them what “good code” is, and you’ll end up in a three-day argument.

That’s not a business — that’s an insiders’ hobby club.

You’re not selling a solution; you’re selling an opinion.

Conclusion: Small Teams Should Stay Away

An AI startup that tries to “optimize software engineering” is really trying to change a company’s culture and power dynamics — which takes influence and access, not just tech.

There are exceptions, of course — take Cursor, for example.

Its genius lies in not trying to fix software engineering at all.

It serves individual developers: helps them code faster, save time, maybe even slack off a bit.

Its value is immediate and tangible, completely sidestepping all those management and ROI traps.

But in the end, that’s just the local optimum of the tech world — a survival trick in an age of over-competition.

If you really want to build something big, step outside the circle.

Use insider knowledge and technical depth — to make money from outsiders.