Creation Is Cheap. Coherence Is Not.

One thing I think AI is going to make much harder for engineering orgs is something we probably don’t talk about enough:

It has never been easier to create a new solution to a local problem.

A team needs a better way to capture knowledge? Build one.

A manual workflow is frustrating and time consuming? Automate it.

An internal tool is missing a capability? Create a new tool.

A process has too much friction? Put an agent in front of it.

Wait. We actually talk about this a ton! And this is the amazing world we live in. In fact, a lot of it is exactly the kind of thing we should want from AI.

What we don't exactly talk about enough is:

AI has changed the economics of creation much faster than it has changed the economics of ownership.

When building something that required any kind of meaningful engineering investment, it created this natural pressure to ask whether the capability already existed, who should own it, or whether it is worth building at all.

That friction is gone.

Now a locally useful solution can go from idea to something fully working quickly enough that building it is way easier than the hassle of discovering what may already exist. This is especially true in larger engineering organizations.

Dozens of individually reasonable decisions can very easily become an unreasonable engineering environment. More tools, more internal platforms, more knowledge stores, more abstractions, and more ways of accomplishing the same task. Oh, and more things that eventually need owners, maintenance, consolidation, or deletion.

And I don't think we should knee-jerk to putting approval gates in front of every idea. That would throw away much of the goodness AI gives us here.

However, I do think leadership needs to become much more intentional about what happens after creation.

We need better discovery and visibility so engineers can see what already exists before building something new.

We need clearer expectations so experiments don’t quietly become permanent and owner-less.

We need to recognize when several local solutions are actually proof we have a real organizational need.

We need a defined path so our experiments can become real, supported capabilities.

We especially need to be much more comfortable deleting things that are no longer useful.

Technical leaders must curate the engineering environment so that local experimentation can happen without allowing the overall system to become messy. It's so much more than just encouraging innovation or defining standards.

AI can help with that too. It can expose duplicate capabilities, connect teams that are solving similar problems, identify stale or conflicting knowledge, and make existing capabilities easier to discover.

Leadership decisions still matter.

What should remain local to a team?

What should become a shared capability?

What solutions should merge?

What can we delete?

Do we really need this?

Those decisions become more and more important as building is becoming cheaper.

This exposes a real challenge for AI-amplified engineering: learning how to let experimentation stay cheap while keeping the organization coherent enough to understand, operate, and evolve.

AI creates abundance. Leadership has to create coherence.