Every in-house platform has one central figure. Even when there is a team of four, there is one person who holds the architecture in their head, makes the real decisions, and keeps the thing alive. When that person jumps to another company, the tool withers — fast. I watched this happen across an entire generation of data infrastructure, and it is about to happen again with agent platforms.
The objection comes up in every enterprise conversation about agent infrastructure: we have strong engineers, why not just build this ourselves? It is a fair question, and the honest answer is — you can, and you will get real value quickly. The data era proved that too. When the old pipeline tools failed, companies everywhere built their own. For a while, it worked.
Then Airflow won. Not because it was pretty or fun to use, but because it was complete — you could not think of anything left to add to it. And complete products eat in-house tools on a predictable schedule. The internal tool stops evolving the day its central figure leaves. The open source or commercial product never stops, because thousands of organizations are pushing on it and a dedicated team wakes up every morning with nothing else to do. The decay is structural, which is why homegrown platforms decay is less a risk to manage than a law to plan around.
There is a second structural problem: your best engineers have a competing mandate. The team that could build your agent orchestration layer is the same team being asked to build AI for your actual business — and they have been given three or six months, not three or six years. Hard enough to believe a startup with twenty million dollars in funding can out-build the labs at this layer. A side project inside an enterprise, on a deadline, with a retention risk at its center? The math does not work.
The hundred largest companies will build internally anyway, and a few of those tools will survive. For everyone else, the right question is not whether your team could build it. They could. The question is what your platform looks like eighteen months from now, after the one person who built it has updated their LinkedIn.
Related Essays
Homegrown Platforms Decay
Internal agent platforms are built by ambitious individuals with other jobs. When those engineers move on, the platform becomes a liability.
What the Build-vs-Buy Data Actually Shows
From Stripe to a five-person startup, the agent stack is mostly blue — built in-house. The harness is bought. The middle of the stack is built. The opportunity sits in turning blue dots green.
Your Real Competitor Is the Internal Build
Enterprise agent shortlists now include a fourth option — paying someone to build it in-house. Flexibility, not features, is what beats that option.
Key takeaways
- Every in-house platform has one central figure, and when that person leaves the tool withers fast — this pattern repeated across an entire generation of data tooling.
- Internal teams get three to six months and a competing mandate, while a dedicated vendor compounds on the problem full time for years.
- Complete products win not by being prettier but by leaving nothing left to add, which is how Airflow eventually ate the in-house pipeline tools.
FAQ
If we have strong engineers, why shouldn't we build our agent infrastructure in-house?
You can, and you will get value fast. But your engineers have a three-to-six-month window and a competing mandate — usually building AI for your actual business. A vendor compounds on the infrastructure problem full time. The honest comparison is your team's sprint against someone else's marathon.
Didn't large companies successfully build their own data platforms?
Many built them, and most abandoned them. In-house data pipeline tools were everywhere until Airflow matured into a complete product and steadily ate them. The largest hundred companies may sustain internal platforms; almost everyone else eventually migrates to commercial or open source options.