Your engineers will build it in secret
Give someone AI and they solve the problem that bugs them. That's how you end up with 18 versions of the same agent.
I think this is super important to realise. When an engineer, or anyone frankly, gets hold of AI, the first thing they tackle is a problem that matters to them, because they finally feel empowered to solve it. Often it’s something that isn’t even part of their day to day.
Cost calculators everywhere
The number of cost calculator agents I’ve seen built in Claude Code recently is frankly amazing.
This is one example of the agent sprawl Zohar talks about. It’s also how you end up with 18 versions of an observability agent, or a chatbot, that all do the same thing.
Guardrails aren’t gatekeeping
Governance and guardrails aren’t meant to be gatekeeping. They can be a way to funnel effort toward the things that actually matter:
- security
- serviceability
- platform awareness
Solid communication and talent management are other areas that come to mind.
The platform gap
Enterprises don’t typically invest in these spaces, the platform, so the new generation of engineers is using AI to solve it for themselves. In a way, that just creates more surface area and more chaos.
I think it’s a problem most will have to face. Either invest in these areas, platform engineering in particular, or your engineers with AI are going to do it for you. In secret.