Companies that deployed AI coding tools without restructuring how their engineering teams operate now face a sobering reality: the promised productivity surge never materialized. Claudio González, CTO of the software consultancy intive, argues that the real winners aren’t chasing better tools. They’re rebuilding their org charts around them.
The shift comes as AI assistants become as routine in development workflows as Stack Overflow once was. Stack Overflow’s 2025 survey shows four out of five developers now use AI on the job. Yet that same data reveals a troubling countertrend. Trust in AI-generated output has declined even as adoption has climbed.
González has seen this pattern repeatedly with clients who buy licenses and wait for results. “Adding a code assistant on top of the old one doesn’t get you there,” he told 150sec. Google’s 2025 DORA report backs this up, describing AI as an amplifier rather than a shortcut. Strong teams get stronger. Chaotic teams get messier.
The underlying problem, González insists, lives in workflow design rather than tool selection. On one major platform build, intive achieved a 40% jump in total development output, 35% faster feature cycles, and 50% faster modernization. None of it happened until the process itself changed completely. A McKinsey analysis from November 2025 reached the same conclusion: pairing generative AI with genuine process redesign pays off far more than bolting an assistant onto yesterday’s pipeline.
**From manual builders to architects of intent**
The old model pushed headcount at problems. Bigger teams, more tickets, more grinding. González sees leverage shifting toward smaller groups with sharper architectural judgment. AI agents handle boilerplate, test generation, and legacy code interpretation. The humans decide what gets built and verify it’s correct.
“The constraint is no longer how many hands you have,” he said. “It’s how good your people are at deciding what to build.”
That logic applies across industries. Retail recommendation engines, logistics platforms, health tech, fintech. Gartner projects AI assistants will surge among enterprise engineers within years, which makes the real question less about adoption and more about who actually belongs in the room.
**The real blocker is rarely the code**
The hardest challenge in software isn’t writing new code. It’s decoding what already exists. Many companies run on systems built decades ago, undocumented and maintained by people long retired. The UK’s National Audit Office reported in early 2025 that the government still operated at least 228 legacy IT systems, with no funded plan to fix roughly half of them.
González rejects the instinct to blame old code. “It’s not the legacy itself. It’s the lost knowledge around it.” His teams deploy AI as a “systems historian,” extracting buried business logic before any rewrite begins.
**Strategy first**
Even with optimism about the tools, González keeps circling back to one limit. Unclear goals, missing early wins, shaky stakeholder confidence. These are organizational failures, not technical ones.
A 2025 MIT study found roughly 95% of corporate generative-AI pilots delivered no measurable return. RAND found over 80% of AI projects fail, roughly double the rate of non-AI IT projects. Most often because leadership misunderstood the problem, not because the technology fell short.
“Strategy still has to come first,” González said. “AI just makes execution survivable.”
