The trap: 'we can't do AI until we modernize'
It's the most expensive sentence in enterprise IT. Modernization programs run years and routinely overrun; meanwhile the manual work — re-keying between systems, reconciling exports, chasing data across portals — burns money every single day. Waiting for the migration to automate is backwards: the manual glue-work between old systems is precisely where automation pays most.
How agents operate systems without APIs
A production agent reaches a legacy system through whichever door exists, in order of preference:
- Real APIs or database access where they exist — fastest and most robust
- File interfaces — the exports, EDI drops, and scheduled reports the system already produces
- The user interface itself — agents that navigate screens the way a trained operator does, with verification at each step
- Email and documents — because half of legacy 'integration' is actually a PDF sent to an inbox
Where this pays off first
The classic wins: order data re-keyed from a portal into the ERP; invoices reconciled between the accounting system and a bank export; inventory synced between an old WMS and a modern storefront; customer records deduplicated across systems that have never spoken. Each of these is usually a person — or three — doing swivel-chair integration all day.
Because the agent sits outside the legacy system, nothing about the system changes: no upgrades, no vendor negotiations, no migration risk. And when you do eventually modernize, the agent's workflow logic moves with you — it's the integration layer, not another dependency.
What to bring to a scoping conversation
Three things: which systems are involved, what the manual work between them looks like (a rough weekly hour count is enough), and where errors hurt. From that we can tell you quickly which bridge is buildable, what the guardrails need to be, and whether the ROI clears the bar. The answer is honest either way — some legacy workflows genuinely aren't worth automating, and knowing that early is worth the call by itself.
Frequently asked questions
Can AI integrate with systems that have no API?
Yes. Agents work through file interfaces (exports, EDI, reports), the user interface itself (operating screens like a trained user, with per-step verification), and email/document flows. An API is the preferred door, not a requirement.
Is automating a legacy system risky?
Less risky than the status quo when built properly: agents operate with the same permissions as a human user, verify each step, log every action, and route anything unexpected to human review. Compare that to unlogged manual re-keying — which is where the errors live today.
Should we wait to automate until after we modernize our systems?
Usually no. The manual work between legacy systems costs money every month, and migrations take years. Automating now pays immediately, and the workflow logic carries over to the new stack — the agent layer becomes your integration layer.