HnyComb
Start free

From Jeremi Karnell · November 2026

A2A Needs a Harness: Agents From Running the Practice to Selling It

Agent-to-agent work will take over more of an advisory practice, from running it day to day to finding the right path and running diligence, but only as fast as each side can trust the other's agent. That trust comes from a harness: governed loops with checkpoints, permission tiers and owner approvals.

Jeremi Karnell · 9 Nov 2026

Agent to agent is where succession and M&A work is heading. A buyer's agent asks, the target's agent answers, and the back-and-forth that used to take weeks of spreadsheets and email happens in hours, with a person approving each step. Today that exchange is narrow: finding what is missing and drafting the answers, after an NDA is in place. It will not stay narrow. The same pattern reaches back into running the practice, years before any sale: a network asking how you compare, a continuity partner keeping an agreement current, a next-generation advisor buying in.

Agents can already talk to each other without rules. The real question is whether anyone will trust them with a firm's real numbers. Two agents discussing a firm's revenue, its clients and its price is exactly the conversation no owner lets run unsupervised. And trust has to run both ways. A seller needs to know what their own agent may do, and also what the buyer's agent is allowed to do with what it receives. That is what a harness provides: limits that are explicit, enforced and visible to the other side. A2A will move only as fast as that trust does.

The third act in a play we have been writing

As far as I can tell, this is the third thing ValuationMCP has done first in this category, and each one set up the next.

Live from Agent to agent →

First came the MCP-native app: a valuation you can run from inside Claude or ChatGPT, instead of a website you have to visit. Your agent can connect to 55 MCP servers to source data directly from the systems of record that run your practice. Then came agent to agent: your agent can talk to a buyer's agent, with a person approving each step, instead of everyone emailing spreadsheets. I wrote about both on LinkedIn; the links are in Sources.

Each step gave agents more reach, and reach is exactly the problem. A2A shipped with approvals built in: two agents, with a person signing off on each step. The harness takes those rules and makes them the foundation of every loop, which is why it is the third act and not an afterthought. It is what makes the first two, and every A2A motion after them, safe to use on real work.

What a harness is, without the jargon

Think of a smart new hire on day one. You would let them read anything in the file. You would welcome their suggestions. You would not hand them the company card on the first morning.

And there are a few things you would never let them do alone, however capable. Send a message to a client in your name. Change what a client pays. Sign the NDA. Decide that the work is finished.

That is a harness. It is not the intelligence. It is the set of rules around the intelligence: what it may look at, what it may suggest, what needs your sign-off, and what it can never do.

Live from Agent harness →

The models will keep getting smarter. That is not the part I worry about. A smarter model with no rules around it is just a faster way to send the wrong thing to the wrong person. The rules are what turn a capable model into something you would let work on a sale.

Four lanes for every action

Every action an agent can take at ValuationMCP sits in one of four lanes, decided before it ever runs. Here's what happens when the agent asks.

Live from Agent harness →

Looking runs on its own. Suggestions land on your desk as a card, and nothing happens until you act on one. Changes run only after you click approve. And some things are off the table entirely: the agent cannot do them however it is asked, and every attempt is written down.

The lanes aren't a policy memo sitting next to the software. They are built into it: every action gets its lane when it is created, and the picture above is drawn from that same list.

Pulls from your systems are worth a sentence of their own. The agent never starts one. It proposes the pull, you start it, and only the totals you approve are saved, each with its source. A buyer's agent never pulls from a target's systems at all. It asks, through a data request, under an NDA.

The model writes the words. The engine owns the numbers.

This is the rule that matters most when the numbers decide who gets paid.

The agent can put the issues in your data in order, phrase the next question, and draft a plan. It does not state, compute, round or restate a figure. Every figure comes from the deterministic valuation engine, shown as a range with its data coverage. Where the model writes text, a sentence with a digit in it is dropped before you see it. Figures appear only as facts the app copies from the engine.

It also cannot grade its own homework. A move is marked verified only by a saved valuation run, never by the agent saying it worked. And drafts stay drafts: the agent prepares the request, a person sends it.

What a loop looks like

An agent run isn't an open-ended chat. It's a short routine that follows the same steps every time, writes down each one, and has firm limits on how long it can run. Here is the one that checks your data before every valuation.

Live from Agent harness →

Two things are worth noticing.

It keeps a record nobody can edit. Every step is written down once, in order, and never changed afterward. If the run is interrupted, it picks up where it left off, and nothing gets done twice. You, your compliance team or a buyer's advisor can read back exactly what happened.

And it knows when to stop and ask. If two sources disagree on your revenue, it doesn't pick one. It asks you, and waits. The decision stays yours; the agent's job is to make sure you're asked the right question at the right time.

Where it runs today

The same harness runs every loop, with the same four lanes and the same rules.

Live from Agent harness →

For an owner, that is the loop that checks the data before any number is computed, and loops that work the operating margin, the client book, billing and growth. Ledger, the hosted agent, works an assigned move on its schedule and leaves every change as a card.

For a deal, it is two loops, one for each side. The buyer's loop finds what is missing and proposes requests. The target's loop drafts answers from its own record. They exchange messages over A2A, nothing is drafted or shared until an NDA is in place, and an incoming message only ever becomes a card for a person to read, never an action.

Where the A2A motions go next

The deal loops running today are the first motions, not the last. Over time, more of the work around a practice becomes something two agents do between them, each under its owner's rules. Much of it starts long before anyone talks about a sale.

Running the practice

Benchmarking with your network. A network or home office asks its affiliates how they compare on margin, growth and client concentration. Today that is a spreadsheet request every quarter. An owner's agent can answer the network's agent from the owner's own record, sharing only the figures the owner approves.

Keeping a continuity agreement live. Most continuity agreements are signed once and forgotten. An owner's agent and the continuity partner's agent can keep one current: the valuation refreshed on schedule, the client book mapped, the terms checked against what the practice is worth now.

Bringing in the next generation. An internal buy-in runs for years. The owner's agent and a junior partner's agent can keep the schedule, the annual valuation and the equity math in step. Every figure comes from the engine, and every change needs both people's sign-off.

Working with lenders and custodians. A lender's agent asks for the financials behind an acquisition loan or a recap. A custodian's agent reconciles billing. Each gets an answer drafted from the record, and none of them gets a pull into your systems.

Doing the deal

Finding the right path. An owner's agent lays out the realistic options against the valuation and what the owner actually wants: an internal successor, a sale to a peer, joining a network. Candidates' agents respond. No name or number leaves the owner's side until the owner says so.

Leading due diligence. The buyer's loop works the checklist and the target's loop answers from its own record. Gaps become requests instead of email threads, and both sides' advisors read the same step log.

Structure and transition. Proposals on terms and earnouts go back and forth, with every figure coming from the engine. After close, the same loops can track the client retention an earnout depends on.

A practice that already runs this way doesn't start diligence from a cold file. The record is current, the step log already exists, and the rules are already set.

Trusting the other side's agent

Every one of those motions runs into the same problem: your agent is governed, but is theirs? Today's answers are structural. Nothing is drafted or shared until an NDA is in place. A buyer's agent can request but never pull. An incoming message only ever becomes a card for a person to read, never an action. The next step is making each side's rules legible to the other, so a seller can see what a buyer's agent is permitted to do before answering it. That, more than any single feature, is what lets A2A take on bigger parts of a deal.

Why this is the real moat

Anyone can wire a model to a data source. In succession and M&A, the failure that ends the product is one confidently wrong number, or one message sent that should not have been.

A harness is how you make that failure structurally unlikely, instead of promising to be careful. I think this category will end up sorting vendors by it.

It is also, frankly, the boring part. Nobody demos a permission lane. Nobody screenshots a step log. But an owner deciding whether to let software near a sale does not need a clever answer; they need to know what the software is not allowed to do, and that the list is enforced by something other than good intentions. A buyer needs the same thing from the other side of the table. So does the M&A advisor sitting between them, who has to explain to both where every figure came from.

MCP-native, then A2A, then the harness. I am not claiming the story is over, only that these were the first three chapters and the order was not an accident.

What it does not do

It is not an appraisal, a fairness opinion, or investment, legal or tax advice. It does not replace your attorney, your CPA or the buyer's diligence. It is a way to let an agent do the legwork while the decisions, and the sending, stay with people.

If you want to see the parts, the harness page walks through them, and the agent-to-agent page shows the deal loops. If you want your firm's own number, run a valuation in Claude and watch where it stops and asks.

Market commentary from public sources, quoted as each publisher released it. Not investment, legal or tax advice, and not an appraisal of any firm.

HnyComb produces educational estimates, not appraisals, fairness opinions, or investment, legal or tax advice.