AI is an over-hyped distraction. AI is giving engineering teams superpowers.
Both are true.
If you drop AI into an existing process, it will break. Everyone must now use Claude Code. You code three times faster, but now your pipeline is overfull and people are learning the best approaches on the fly. Seniors in the team are inundated with Pull Requests of widely varying quality and relevance[1]. This is pure frustration.
Counter that with a team that has renovated their SDLC from the ground up. Many PRs might not need review at all. Engineers collaborate on shared approaches and skills. Non-engineers are prompting and shipping changes themselves[2]. These teams are getting genuine lift.
Before AI, a team with a messy codebase and resource-constrained review process could still ship and even ship well. The human pace of code generation simply didn't stress the structure to breaking point. AI removed that governor[3].
Not only are both of these true, both are playing out under the bright, overly-amplified lights of social media.
The Cost of Transformation
The real cost of introducing AI tools is a significant overhaul of every process, person, and team it touches. This cost is often not recognized by those pushing AI into the organization.
Layoffs at tech companies are being branded as "AI washing". I'm sure this is true. It's also equally true that organizational transformation is hard. Very hard. And reducing the team is one insanely-blunt lever to making it easier.
The Software Development Lifecycle (SDLC) is deeply entrenched -- systems, processes, tools, institutional knowledge. Right now it's being upended. Many engineering teams are trying to pick up the pieces without the resources to do so[4].
The teams I've seen succeed are investing heavily -- restructured codebases, risk-tiered reviews, faster feedback loops, better inputs, radically changed workflows. These aren't incremental tweaks. They're a reboot[5].
The 2x question
I spend half my time on renovating our process. Ergo, I need at least a 2x uplift from AI tooling to break even. Equally, if I exceed 2x I start to see a compounding benefit.
If you're in a situation where AI can't ~2x your particular process or workflow, then it simply does not make sense at the moment. You are better off being a fast-follower and picking up the lessons at a tenth of the cost.
If you can get there (and I believe most can), then the benefits can be staggering. But it's not free and certainly not the cost of expensing a bunch of Claude licenses.
The challenge is it's leadership issuing AI mandates and it's leadership that isn't making the investments.