Sometimes none of the frameworks fit. Rolling your own is the right call rarely — and the wrong call often. Here's the first-principles test, and the cashflow guardrail.
"Build your own agent framework" is a tempting answer to framework frustration. It's sometimes right and usually wrong. The test is first-principles — what do you actually need that the frameworks don't provide — with a cashflow guardrail on the cost of building it.
First Principles: What You Actually Need
Strip it down: an agent framework gives you a loop, state, tools, and orchestration. The question is which of those you need that the frameworks don't provide — specifically, not generically. "More control" is not an answer; "the framework's state model can't represent X" is.
- A loop the frameworks don't support.
- A state model the frameworks can't represent.
- A performance profile the frameworks can't hit.
- An integration the frameworks don't allow.
The Cashflow Guardrail
Building a framework is expensive — in time, in maintenance, in the cost of every feature the frameworks would have given you for free. The guardrail: the cost of building and maintaining your own must be lower than the cost of working around the framework's limits. Usually it isn't.
- Cost of building + maintaining your own.
- Cost of working around the framework's limits.
- The first must be lower than the second.
When Rolling Your Own Wins
It wins when your workflow's needs are specific and stable — a narrow domain, a fixed set of tools, a state model the frameworks genuinely can't represent — and the maintenance cost is bounded. It loses when the needs are vague and evolving, which is most of the time.
What to Refuse
- Rolling your own because the frameworks have rough edges (they all do).
- Rolling your own without a specific, stable need.
- Rolling your own without costing the maintenance.
- Rolling your own before prototyping in two existing frameworks.
Conclusion
Rolling your own agent framework is rarely right and sometimes necessary. The test is a specific, stable need the frameworks can't meet — with a cashflow guardrail that the build-and-maintain cost is lower than the workaround cost. If you can't pass the test, use a framework.
About FACTA
FACTA helps startups and growth-stage teams turn AI into production systems that keep running — not demos that impress once.
We design the architecture around the parts that actually break under real usage: tooling you own, credentials you control, failover, cost controls, observability. The boring infrastructure that keeps a system alive after launch.
Led by Matías Baglieri and Carolina Fogliato, we focus on one thing:
AI leadership that builds. Not just advises.
Tell us what the frameworks can't do for you.
We'll tell you whether that's a roll-your-own case or a workaround case. See the framework comparison for what's available.
Explore AI Automation