‘Build or buy?’ is still relevant

In recent months, the conversation has shifted from “AI can help build software” to the idea that generative AI may render entire categories of SaaS obsolete.

The logic often sounds like this: if a model can reliably map complex workflows — even legacy systems like Cobalt — and generate the code you need on demand, why maintain the surrounding tooling at all? Why continue purchasing software from vendors? Why not simply generate what you need?

Wouldn’t that be nice.

Generative AI does meaningfully lower the barrier to creating software. Prototypes can be produced in minutes. Workflows can be documented instantly. Or seem to be, at least.

But lowering the barrier to creation is not the same as eliminating the burden of ownership.

Mapping a workflow is not the same as governing it.
Generating code is not the same as validating it.
Vibe coding a system is not the same as sustaining it over time.

When people point to Cobalt as something a model can now “handle,” I find myself pausing. Cobalt sits behind systems where precision, auditability, and stability matter, including significant parts of financial infrastructure. Reproducing or translating those workflows isn’t simply a technical exercise. It’s a question of verification, compliance, and accountability.

If AI generates the workflow, who signs off on its correctness?
Who stress-tests edge cases?
Who ensures regulatory alignment?
Who is responsible (or notices!) when something fails?

I recently saw someone writing about building their own version of a personal finance app — something like Rocket Money — in an afternoon. Wild that’s now possible, but it raised a different question for me: why would anyone choose to maintain and secure a tool that directly accesses your financial accounts when there are dedicated companies structured to handle that responsibility?

Yes, you can build it.

But now you own the integrations.
You own the security updates.
You own the API changes.
You own what happens when something breaks.

The same dynamic exists at enterprise scale. The drafting layer may get lighter, while the oversight layer gets heavier.

The claim that SaaS becomes obsolete assumes that software’s primary challenge is initial construction. In practice, much of the cost and complexity lives in maintenance, monitoring, updates, security, and governance.

AI compresses creation. It does not remove ownership.