Build
Build vs buy an AI employee
Tobiloba Odejinmi · 1 Feb 2026 · 6 min · 1,241 words
Direct answer
Build when the process is specific to how you already work and you need it plugged into tools you already use. Buy when a product already does that job inside those tools and you can turn it on without a new religion. A one-process build is $5,000. The full week is $7,500. Several workflows start at $15,000. If a vendor needs a quarter and a new login to take the first pile off your team, you are not buying an employee. You are buying a programme.
- Name the process before you name the vendor.
- A new login is a cost, not a feature.
- $5k, $7.5k, and $15k+ are for live workflows, not decks.
- The first call should end with hours, miss-cost, and a yes or a no.
Should you build or buy an AI employee?
Start with the process, not the market map. If you cannot name the pile, buying will give you software and the same pile. Building will give you a project and the same pile. I will not help you shop until the hours and the exits are on a page. That page is days 1 to 3 of my week even when the answer is 'buy the thing you already have.'
Build when the work is yours: your refund exceptions, your hiring bar, your follow-up note, your internal copy between two systems. Buy when the work is generic and already solved inside a tool you will not leave. The honest cases are mixed. Most companies already paid for a help desk. They have not paid for their process to be written down.
When is buying the faster path?
When the vendor lives where the work already lives, the gates exist as settings, and a person on your team can turn it on before lunch. Macros with a model behind them. A CRM first-touch that writes on the lead. A screening view inside the ATS. Those can be faster than a week, and I will say that on the call.
Buying is not faster when the quote includes discovery, professional services, and a new workspace. That is a build with their logo. Price it like a build. Ask for one live process and a handover. If they cannot do that, you are in a programme. Programmes have their place. They are not a 7-day sprint.
When does a platform become a new religion?
When usefulness requires your team to live in the vendor's inbox, ticket list, or 'agent console' instead of the tools they already open. When knowledge has to be re-uploaded into a store you cannot export. When escalation is a transcript in their UI and your people have to retype the decision into the real system. That is a new religion. I will not baptise you.
I have a simple test from Zeeh and from every diligence I have sat in. Can another team plug in before lunch. Can you explain the cost and the logs. If only a champion understands production, that is a risk. Fancy agent slides do not survive that meeting.
What does a $5k, $7.5k, or $15k+ week buy you?
Five thousand dollars is one workflow: we pick the process, I build it into the tools you already have, we test it, I write the page, we do the call. Seventy-five hundred is the week most people book: that, plus a plan for the next process, a knowledge store if you need one, proper API connections, monitoring, and 30 days of support. Fifteen thousand and up is more than one live workflow, deeper integrations, and someone owning the architecture after launch.
None of those prices is a licence for unlimited agents. They are weeks of work with a finish line. If a vendor's annual seat cost is lower, still ask what is live in seven days. A cheap seat that does not take the pile off your team is not cheaper. It is a quieter invoice.
How do you compare a vendor to a one-process build?
Put the same columns on one page. Days to first live process. Tools the team must open. Who owns the map. What a handoff contains. Whether hard gates are permissions or prose. What week two support looks like. Where the logs live. Who can pause it. If the vendor cannot fill those columns, they are selling a feeling.
Then put the pile next to the price. Hours per week, miss-cost, and whether a person still owns 10 to 15 percent. A build that removes a real pile at $7,500 is easy to defend. A platform that adds a weekly steering meeting is easy to defend in a different way, until someone asks what got faster. I have been in those rooms. The question is unglamorous. It is the right one.
What question should you ask on the first call?
What should we take off your plate, what would that save, and what would it take to ship. That is the whole call. Thirty minutes. If the answer is a platform tour, we are in the wrong meeting. If the answer is a pile with examples, we can talk about day 1. If there is nothing worth doing, I will say so. I have enough work. I do not need to sell you a week that will embarrass both of us on day 7.
Bring the doer. Bring access questions. Leave the vision deck. I still write the code. I still want the week to end with something running in the tools you already use. Build or buy, that is the bar. No new login if we can help it. No new religion either.
What if you already started a vendor trial?
Keep it if it already writes into the queue you run. Kill it if the trial is a second inbox your team checks out of guilt. I have watched trials expire with nobody able to say what got faster. That is not a procurement failure. That is a process you never picked.
If the trial is halfway useful, we can treat it as a connection. Days 4 to 6 of a build can sit on top of a tool you already bought. The employee is still your map, your gates, and your handover. The vendor is a pipe. Pipes are fine. Pipes that demand worship are not. Start with one process. Price it like a week: $5,000, $7,500, or $15,000+ when you already know you want more than one.
Questions people ask
When do you tell someone to buy?
When their help desk or CRM already has a first-pass feature that hits the same exits, and the team will actually use it. I am not here to rebuild a button they already have.
When do you tell someone not to buy a platform?
When the pitch starts with agents and ends with a new workplace. If the pile would still sit in the old tools, the platform is a tour.
Is the week cheaper than a vendor?
Sometimes in year one. Sometimes not. The question is time-to-first-live-process and whether you still own the map. Seat prices hide that.
Can we mix build and buy?
Yes. Buy the connection you already trust. Build the workflow the vendor will not touch because it is yours. That mix is normal.
What should we bring to the first call?
A pile, a tool list, and someone who does the work. Thirty minutes. If there is nothing worth doing, I will say so.
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.
