Custom vs Off-the-Shelf AI Agents
Custom is right less often than the people selling custom will tell you. Here is the boundary, and it is usually 80/20.
Side by side
| Off-the-shelf tool | Custom-built agent | |
|---|---|---|
| Cost | Published, usually per seat or per usage unit | A build, plus a maintenance tail that lasts as long as you use it |
| Live in | The afternoon | Weeks |
| Fits your process | Your process bends to it | It bends to your process |
| Who maintains it | The vendor, included | You or whoever built it, forever |
| When the vendor pivots | You migrate, on their schedule | Nothing happens |
| Integrates with your stack | Whatever they support | Whatever you pay for |
| Worth it when | Your process is standard, which it usually is | Your process is genuinely non-standard, and that is the source of your advantage |
The 80/20 rule, said plainly
For most companies, an off-the-shelf tool does 80% of the job today, and a custom build does 100% of it in two months. The 80% starting this afternoon is almost always worth more than the 100% starting in July.
We build custom agents and we still tell people this, because the alternative — taking money for a build that a $40/month tool would have covered — produces a client who cancels in month four and tells people why.
The test
Describe, in one sentence, what makes your version of this process different from everyone else’s.
If the sentence comes easily and the difference is the reason customers choose you, custom is defensible. If you find yourself saying “well, we do it a bit differently” and then trailing off, your process is standard, and standard processes have tools.
What custom actually commits you to
Not a purchase — an obligation. Custom code has a maintenance tail that lasts exactly as long as you use it. An API changes, a model deprecates, a field gets renamed, a schema shifts. Somebody handles that, forever, and if nobody has been named then the answer is that it degrades quietly until it has been wrong for a while.
The vendor of an off-the-shelf tool carries that cost for you and spreads it across every customer. That is genuinely the best argument for buying, and it is the one custom shops mention least.
The shape that usually wins
Buy the tool for the standard part. Build only the specific step where your process genuinely diverges — and connect the two. That keeps the maintenance surface small, which is the thing that actually determines whether an automation is still running in eighteen months.
Before committing either way, it is worth seeing what the tools in this market actually charge. We compiled that: the AI Agent Pricing Index lists what 26 vendors publish, including the seven who publish nothing at all.
When off-the-shelf tool is the right answer — and we would rather say so than sell you something that will not work:
- — Your process is normal. Invoicing, scheduling, support triage, lead routing — these are solved, and a tool that does 80% of it today beats a custom build that does 100% of it in two months.
- — You cannot describe what makes your version different in one sentence. If you cannot, it is not different.
- — The volume does not justify a maintenance tail. Custom code is a standing obligation, not a purchase.
- — You are pre-product-market-fit and the process will be unrecognisable in six months.