Industry
Selling a company that runs on agents
Tobiloba Odejinmi · 17 Jun 2026 · 6 min · 931 words
Direct answer
If you sell a company that runs on agents, the buyer will not open your vision deck first. They will open cost, uptime, and whether someone besides you can explain the database. We sold Insurpass after more than 100 companies and a team of 17. We sold Shopl after Techstars '22. Fancy AI slides do not survive that meeting if the books and the logs are a mess.
- Diligence is a tour of the boring systems, not the agent demo.
- Every workflow needs an owner, a log, and a way to run without you in the room.
- Cost and uptime will be modeled. Surprise spend is a price chip.
- Write the runbooks before the data room. You will not write them well under a clock.
They will ask unglamorous questions
We sold Insurpass after growing it to more than 100 companies and a team of 17. The conversations that mattered were not about slogans. They were about whether the product would still run after the wire hit.
If you later add AI employees to that stack, the same questions apply. Who owns the workflow. What happens when it is wrong. Where are the logs. I would rather you write that down this quarter than invent it in a data room.
Agents look like people until diligence
Inside the company, an agent can feel like headcount you do not have to manage. In diligence it looks like software with opinions. Software needs owners, versions, costs, and a way to turn it off.
If your answer is a Slack mention of one engineer, the buyer will discount the whole operating model. They are not buying your overtime. They are buying a machine they can keep.
Cost will be modeled whether you like it or not
Buyers open cost. They will take your model bill, your monitoring, your retries, and your human fallback, and they will annualize it. If you have never added those numbers up, they will add them up for you, with worse assumptions.
I keep process prices visible when I build. A workflow that costs a known amount is an asset. A workflow that 'depends' is a conversation you will lose.
- What it costs in a quiet month and a busy month.
- What breaks when a vendor raises prices.
- Which jobs a person still does, and what that salary is.
- Whether success makes it cheaper or more expensive.
Uptime is a story about Monday
They will ask how often it breaks. Have the incidents. Have the minutes. Have the person who was on call. A chart that says 99 and something is not a substitute for 'here is the last time a customer felt it'.
At 10mg I treat downtime as a patient problem. A buyer of an agent-heavy company will treat downtime as a transfer problem. Same discipline. Different Friday.
Explainable systems sell. Heroes do not.
If only one person understands production, that is a risk, not a flex. I still write the code. I also refuse to be the only one who can explain it.
Shopl went through Techstars in 2022 and we sold it. Nobody in a real diligence thread traded that for a missing schema. They wanted to know the work would continue. Write things down before someone asks.
Build the company you would buy
Name an owner for every agent. Keep logs. Keep a kill switch. Keep the output in systems a new lead can open in the first week. That is how you sell a company that runs on agents without apologizing for them.
If you want a shorter version: make the magic look like software. Software they already know how to diligence.
What I would put in the data room this month
A one-page map of every agent: trigger, tools, owner, cost, last incident, kill switch. A second page with who can explain the database if I am not in the country. A third with the last quarter of uptime the customer actually felt, not the number you put on a status page.
We grew Insurpass with a team of 17. The sale was not a reward for being busy. It was a judgment that the thing would keep running. Shopl went through Techstars in 2022. That is a line on a site. The line that mattered in the room was still whether a stranger could keep the product alive.
Write it now. You will not write it well under a clock, and you will not like the price they put on the missing pages.
Questions people ask
What did buyers actually ask at Insurpass?
What does it cost to run, how often does it break, and who can change it without you in the room. They wanted to know the product would still run after the wire hit.
Do agents help or hurt a sale?
They help if they are just another explained system. They hurt if they are magic only one person can restart. Magic is a risk line.
What about Shopl?
We went through Techstars in 2022 and later sold. The accelerator is a story. Diligence is still cost, uptime, and explainable systems. The story does not replace the logs.
Should I pause automation before a process?
No. Make it duller. Name owners. Write costs. If you turn it off to hide it, they will still ask how the work gets done, and you will look worse.
What should I put in the data room?
A list of workflows, who owns them, what they cost, how you watch them, and the last three incidents. If you cannot produce that in a week, start now.
Written by
Tobiloba Odejinmi
Head of Engineering at 10mg Health. I have run engineering at Zeeh Africa and sold Insurpass and Shopl. I still write the code. If you have one process that still runs on people copying things, we can look at it in thirty minutes.
