June 10, 2026

Agentic Workflows in M&A: What to Automate and What to Leave Alone

"Agent" is the word of the year. In deal work the real question isn't whether to use them, but exactly where they help and where a human has to stay in the loop.

"Agent" has become the word of the year. Every tool claims to have one. In deal work, though, the interesting question was never whether to use agents. It's where they actually help, and where handing them the wheel is a mistake.

An agent is just software that can plan and carry out a multi-step task using tools, rather than answering a single question. Give it a goal and it will break the goal into steps, call other systems, and work toward a result. That's genuinely useful. It's also exactly why you have to be careful about what you point it at. The thing that makes an agent powerful, that it acts without asking at each step, is the same thing that makes it risky. A chatbot that's wrong gives you a bad sentence you can ignore. An agent that's wrong takes a bad action, then takes three more on top of it before anyone looks. The errors don't sit still and wait for review. They compound, because each step builds on the last one's output, mistakes included.

Where Agents Earn Their Keep

The best place for an agent is repetitive work with a clear, checkable result. Gathering background on a company. Pulling together what's publicly known about an owner. Enriching a list with contact details. Drafting a first version of an outreach message. Watching for a signal like a leadership change or a new filing, and flagging it.

These tasks share two traits: they're tedious enough that people cut corners on them, and the output is easy to verify. That second trait is the one that matters, and it's easy to overlook. Verification has to be cheaper than doing the work yourself, or the agent gives you nothing. Enriching a contact passes that test: you can confirm a phone number in seconds. Drafting an outreach note passes it: you read three sentences and you know. The leverage isn't that the agent is smarter than you. It's that checking its work costs a fraction of producing it, so one person can supervise the output of a hundred. The moment verifying takes as long as doing, the agent stops being help and starts being a thing you have to babysit.

Where to Keep a Human

The opposite case is anything that requires judgment, carries real stakes, or is hard to undo. Deciding what a business is worth. Choosing whether to pursue a deal at all. The actual relationship with an owner. These are not busywork. They're the job.

It's tempting to automate them anyway, because an agent will happily produce a confident answer. But a confident answer to a judgment question is not the same as a good one. The deeper reason to keep a human here isn't accuracy, it's that these calls don't have a checkable result. There's no source to point the agent at for "is this business worth pursuing," because the answer lives in your read of the owner, your appetite for the messiness, your sense of where the industry is headed. That's not data the agent has, and it can't retrieve its way to it. A valuation it produces looks exactly as polished as one a partner sweated over, which is precisely the problem: you can't tell them apart by looking, and the cost of being wrong here is measured in deals, not minutes.

The Supervision Line

The cleanest rule we've found: an agent can do anything inside the building, but a human approves anything that leaves it. Research, drafting, organizing, monitoring: let the agent run. Sending an email, committing to a meeting, anything an owner will actually see: that gets a human's sign-off.

This keeps the speed where speed is safe and keeps a person on the irreversible actions. The line is reversibility, not difficulty. A bad draft costs nothing because you delete it. A bad email to an owner you've been circling for a year can't be unsent, and it burns a relationship you can't easily rebuild. So the test isn't "can the agent do this well," it's "what happens if it does this wrong." Anything you can throw away, let it run. Anything an owner remembers, a human signs off. Drawn that way, the agent's mistakes stay cheap, because the expensive ones never make it out of the building.

Why This Matters Right Now

Agents have crossed the line from demo to genuinely useful, which is exactly when they get dangerous. Capable enough to trust with real work, not so capable that they should be trusted with everything. The design choices, what to automate, where to require approval, matter more than the model.

That's how we think about it at Equitle. Agents do the busywork that used to eat your week, the research, the enrichment, the first drafts, the watching, and you keep the judgment that actually wins deals. The point isn't to replace the dealmaker. It's to give them their time back.

Shrey Parekh

CTO, Co-Founder, Equitle