the leaps from gpt-3 to 3.5 to 4 felt transformative, but gpt-4.5 was only a small improvement, with costs rising enough to make companies question if the gain was worth it. progress is still happening, but the giant jumps seem to be slowing.

a mit study showed that 95% of ai pilots inside companies are failing, with only 5% accelerating revenue. the research highlights that buying ai tools has higher success rates than building them in-house, despite the current obsession with developing proprietary solutions.

a second study, measuring the impact of early-2025 ai on experienced open-source developer productivity, found that ai slowed developers down by 19%, even though they believed it sped them up by 20%. prompting, reviewing and retrying created more friction than benefit, turning what seemed like efficiency into delay.

on the business side, meta froze its ai hiring spree, a signal of more caution ahead. many ai startups remain unprofitable, and securing future funding will be harder without clear paths to profitability — echoes of the dot-com bubble are hard to miss.

the hype has also distorted the job market. some suggest ai could replace junior developers, an idea criticized by industry leaders. amazon’s cloud chief called it the dumbest things i’ve ever heard”, noting that junior roles are vital for developing future senior engineers.

ai itself isn’t the problem. it’s powerful, and it will stay with us. the issue is how it’s used: companies chase the “latest technology” without asking if it truly impacts their business or workflows. like the dot-com era, many will fail, but those who align ai with real purpose and discipline will remain.