The MSA Problem: The Real Reason Agencies Won't Sign With Boutique AI Agencies

I have been talking to agencies lately, the big network kind and the large independents, about working together. The conversations start well. The interest is real, the work we do is the work their clients are asking them for, and the fit looks obvious. Then the conversation reaches legal, and it dies. The reason given is always some version of the same sentence: we cannot get you through our MSA process because of security concerns.

An MSA is a master services agreement, the standing contract that has to exist before a vendor can do any work at all. No signed MSA, no purchase order, no project, no exceptions. It is the front door of every large agency relationship. And for boutique AI agencies like mine, that door is staying shut, with security named as the reason. This post is about whether security is the real reason, because I do not think it is.

Let me be fair before I am blunt, because I spent years inside this world. I ran search marketing for pharmaceutical brands, where vendor security review is not theater. Regulated clients have real obligations. A data processing agreement that spells out how client data is handled, access controls that limit who and what can touch a system, audit rights, breach notification duties: these are normal, they are necessary, and any serious boutique should expect them and sign them.

We do. Security is one of the things Thanos Labs sells. Our agents run on least privilege access, read only scopes where reading is all the job needs, isolated environments per client, and a human approval gate on every meaningful action, with logs an auditor can read. If an agency's security team wants to test that, we will open the hood. That is not the part of this story that fails.

A real security objection behaves a certain way: it names a risk, and a control settles it. What we keep meeting behaves differently. Watch for these patterns, because every boutique AI agency I know has seen at least one of them.

• The objection survives every control offered. No client data access at all? Still no. Isolated tenant, read only scopes, human approval on every output, full audit trail? Still no. When no control changes the answer, the control was never the question.
• The questionnaire is welcomed and the MSA never gets papered. Security forms get filled out, calls get held, everyone is polite, and then the agreement itself simply never arrives. A review with no finish line is not a review. It is a waiting room.
• The terms are written so no small firm can sign them. Unlimited indemnity, which means uncapped liability for anything that goes wrong. Insurance minimums sized for a holding company balance sheet. Intellectual property clauses that assign not just the work product but the vendor's pre existing platform, the very asset the agency is trying to rent. The same agency signs comparable risk with large vendors every week. The terms are not a standard. They are a filter.
• The security standard moves. Each answered question produces a new category of concern, and the category never closes. Real security teams work from a checklist with an end. A door that never opens is not a checklist.

Here is the structural problem nobody puts in the email. Agencies sell hours. Their rate cards, their staffing models, their margins, and a good share of their packaged service offerings are all built on the price of a person's time. A project an agency scopes at 300 hours is a 300 hour invoice. An AI agent workforce does that same project in 100 to 200 hours of human supervised work, and the honest price drops with it.

So put yourself in the agency's chair. Signing a boutique AI agency does not just add a vendor. It shows the client the new cost curve. The first time a client watches a 300 hour deliverable land in 120 hours, every other line on the rate card gets repriced in that client's head. The MSA wall does not protect the client from a risky vendor. It protects the agency's book from a comparison it cannot win.

The industry knows the ground is moving. In a Digiday+ Research survey of agency professionals, the share naming AI disruption as their top challenge jumped from 11 percent in 2025 to 38 percent in 2026. And S4 Capital's Sir Martin Sorrell has told analysts plainly that the billable hour is dying, moving his own group toward subscription fees that bundle senior talent with agentic workflows. When the largest players are publicly repricing their own model, believe them. The quiet MSA refusals aimed at small AI shops are the same story told privately.

None of this argues for skipping security review. It argues for an honest one. Paper the real risks: a data processing agreement with teeth, least privilege access, human approval gates, audit logs, breach notification. Give the security review a checklist with an end, the same end every other vendor category gets. Write MSA terms a well run small firm can actually sign, with liability and insurance proportionate to the engagement instead of proportionate to a holding company.

And then price the work the way the work now works: outcomes instead of hours. The agencies that do this will keep their clients, because they will be the ones who brought the client the savings instead of the ones the client caught hiding them. The boutique AI agency is not the threat to that relationship. We are the cheapest way for an agency to get to the new cost curve before a competitor shows the client first.

If security were the issue, controls would settle it. We offer the controls, in writing, and the answer does not move. What moves the answer is the invoice: 300 hours defended against 100 to 200, a rate card defended against a cost curve. That is a business decision, and agencies are entitled to make it. They are not entitled to call it security.

So here is our standing offer to any agency reading this. Send the real checklist. We will complete the real review, sign the real data agreement, run read only, and keep a human on every approval. If we pass, paper the MSA. And if the answer is still no, say the true reason out loud, so both of us can stop spending hours neither of us bills for.