Has the Cost of Building Software Just Dropped 90%?
Agentic AI coding tools are slashing software development costs by automating labor-intensive tasks, unleashing massive latent demand for custom applications, and fundamentally changing what developers do.
After nearly 20 years building software professionally, I’ve witnessed major shifts: the birth of SaaS, mobile’s dominance, blockchain hype, and low-code promises. But nothing compares to what’s happening now with agentic coding. The economics have changed dramatically, and 2026 will catch many off guard.
The Labor Cost Collapse
Consider a typical internal tool project. Previously, you’d need a team handling CI/CD setup, data access patterns, core services, CRUD pages, dashboards, and testing. Add coordination overhead: standups, ticket management, code reviews, handoffs between frontend and backend developers. A month-long project was standard.
Now? Agentic coding CLIs complete the same work in hours. I’ve watched Claude Code generate 300+ unit and integration tests for a complex internal tool in a few hours—work that would take experienced developers days.
The thinking time remains the same. The implementation time collapsed.
Projects that took a month now take a week. With smaller teams, you get the inverse of Brooks’s Law: communication overhead disappears instead of scaling with headcount. A handful of people achieve an order of magnitude more output.
Unleashing Latent Demand
This isn’t just bad news for developers—economics suggests otherwise.
Jevons Paradox explains why: when production costs drop, we don’t just do the same work for less money. We do dramatically more work. Electric lighting didn’t just replace candles cheaper—it created vastly more artificial light overall.
Every organization has hundreds of Excel sheets tracking business processes that would work better as SaaS applications. At $50,000 for custom development, only essential projects get built. At $5,000 (skilled developer plus AI tooling), suddenly far more projects become viable.
The latent demand for software is enormous.
Domain Knowledge Becomes the Moat
Human oversight remains crucial. Pure autonomous coding creates messes quickly, but human-guided agents produce high-quality software fast.
Developers who master these tools become incredibly effective at solving business problems. Their domain knowledge becomes a huge lever—knowing the right architectural decisions, frameworks, and libraries for each project.
The mythical 10x engineer is here. Even better: pairing business domain experts with motivated developers and these tools creates an incredibly powerful combination. Instead of squads with multiple developers, we’ll see tighter pairings of two people iterating rapidly.
Software becomes almost disposable. Bad direction? Throw it away and restart using those learnings. The hard work is conceptual thinking, not typing.
Don’t Get Caught Off Guard
The models keep improving rapidly—faster than benchmarks capture. Opus 4.5 handles 10-20 minute sessions without going off track. Hundreds of billions in GPU investment are just starting to show results.
Yet many software engineers fight this change with familiar objections: “LLMs make too many mistakes,” “It can’t understand my framework,” “It doesn’t save time.”
These assertions are rapidly becoming false. They remind me of desktop engineers dismissing the iPhone in 2007. We know how that ended.
Large corporations remain behind the curve, lost in vendor approval bureaucracy. But smaller companies using these tools gain massive competitive advantages.
One common objection: “LLMs only work on greenfield projects.” Wrong. I’d rather inherit a repository written with an agent and good engineer oversight than one written by a questionable contractor three years ago—no tests, spaghetti code, everyone who understood it long gone.
Agents excel at explaining existing code, finding bugs, and suggesting fixes.
The Shift Is Here
Your job will change, but software always changes. This time it’s happening faster than anyone anticipates. Engineers need to lean into this transformation now.
The cost of building software has dropped 90%. The question isn’t whether this will reshape the industry—it’s whether you’ll adapt quickly enough to benefit from it.