It’s Time to Revisit No Silver Bullet

In 1986, Fred Brooks wrote his seminal paper No Silver Bullet, arguing that no single breakthrough would give us a 10x improvement in software productivity, reliability and simplicity.

His argument was structural: every proposed breakthrough improved only part of the process, while the essential work remained. So the total gain was capped by everything the breakthrough didn’t touch.

AI is the first technology that isn’t a part. It touches every step of building real software: ideation, specification, design, prototyping, coding, validation, debugging.

AI also reaches the part Brooks felt was untouchable: “I believe the hard part of building software to be the specification, design, and testing of this conceptual construct, not the labor of representing it…” And: “If this is true, building software will always be hard. There is inherently no silver bullet.” Brooks also understood that you can’t simply think your way to the right specification. You discover what the right conceptual construct is partly by trying the wrong ones. AI radically changes the cost of that process: building prototypes and trying out ideas is now cheap, and it’ll only get cheaper.

Brooks addresses AI head-on: “Many people expect advances in artificial intelligence to provide the revolutionary breakthrough that will give order-of-magnitude gains in software productivity and quality.” He then turns to Parnas’ argument: AI was problem-specific. There wasn’t a general-purpose AI technology. Parnas was right about 1986 AI. But without meaning to, he was describing one property a silver bullet would need: generality. Not another problem-specific technique, but something transferable across problems. That’s exactly what current AI is. The same technology works across the entire process, including the work Brooks considered essential.

I don’t fault Brooks or Parnas for not anticipating anything like current AI in 1986. Even two years ago, I didn’t see this coming either.

I’d mark November 2025, with Opus 4.5, as when modern AI really came online for software companies. That’s less than a year. AI-first companies, built from the beginning with the ambition of much larger ones, barely exist yet.

At Alt we’re two people. Pre-AI, I believe building a company with this scope would have taken six to eight. That’s 3–4x in the first year.

Is it 10x? No, I don’t think it is.

But is it a big enough jump, this early, across even the essential work Brooks thought would remain hard, that we should revisit his thesis?

Yes. I think it is.