The reality check: productivity promise falls flat
But a new reality check has arrived. According to Bain & Company’s Technology Report 2025, generative AI in software development is far from the transformative leap its backers advertised. While two-thirds of software firms have rolled out AI assistants, adoption among developers remains low. Productivity boosts hover around 10 to 15 percent, and in many cases, the savings don’t translate into measurable returns.
The report notes that simply accelerating code writing won’t reduce time to market, since coding and testing make up only about a third of the full development process. Without redesigning the entire workflow, AI’s role risks being more gimmick than game-changer.
When AI slows developers down
Even more surprising, other research points to AI assistants actively hindering productivity. The nonprofit Model Evaluation & Threat Research (METR) found that developers using AI tools sometimes took 19 percent longer to finish tasks, spending extra time fixing errors or chasing misleading “hallucinations.”
Stack Overflow’s latest survey echoed this frustration. While more developers are experimenting with AI, their trust in these tools has dropped sharply, with many reporting “almost right, but not quite” solutions that still require manual rewrites.
Agentic AI: the new buzzword
Despite the setbacks, Bain flags the rise of “agentic AI” — tools that act autonomously rather than waiting for step-by-step human prompts. In theory, these systems could execute entire workflows. In practice, early tests have been rocky. A much-hyped AI “engineer” called Devin completed only three out of 20 tasks in independent trials. Gartner now predicts that nearly half of agentic AI projects could be scrapped by 2027. The gulf between investor hype and practical returns has left analysts uneasy. Companies are spending billions chasing efficiency gains, but without clear productivity metrics, many of those investments remain stuck in pilot projects. Bain’s report stresses that true value will only come if firms reinvent the entire software development lifecycle — from planning to maintenance — around AI, not just coding shortcuts.
Coding as the new literacy
Meanwhile, other leaders such as Google Brain co-founder Andrew Ng argue against dismissing traditional coding skills. He calls it “some of the worst career advice ever given” to suggest coding will vanish. Instead, Ng sees programming as the new literacy of the AI era — the skill that allows humans to give precise instructions, rather than vague prompts, to machines.
For teenagers hearing Wang’s rallying cry to spend thousands of hours “vibe coding,” the Bain report is a sobering counterpoint. Yes, AI may eventually reshape software development, but the revolution is slower and more complicated than promised. Without systemic changes, coding with AI risks being more about vibes than value.
