meirlabs
Meir RosenscheinBuild AI In-House, Hire a Consultant, or Use an Adoption Pod?
Most small and mid-sized companies without a machine-learning team shouldn't hire an AI engineer yet, or buy a six-month agency project. Start with an embedded adoption pod (your own contact inside the company plus one AI specialist) that ships one real workflow into production in about 6 to 8 weeks. Build a team in-house later, once you have three or four working tools and know what you'll need to maintain.Launch offer: Early clients get 50% off their first build, so your real cost is about half these figures. Book a free AI plan to lock it in.
Most owners frame this as build versus buy: build a tool with your own engineers, or buy one off the shelf. For a company without a machine-learning team, that's the wrong frame. Any of the three paths below can produce a working AI tool. What decides the outcome is which delivery model gets that tool into daily use before the momentum dies. The wrong one leaves you paying for something nobody opens. MIT found roughly 95% of enterprise generative AI pilots stall without measurable impact, mostly on integration and adoption, not model quality. So the real comparison is speed to a used tool.
What are your three options?
There are only three. First, hire in-house: bring on an AI engineer or a small team, put them on payroll, and build tools yourself. Second, hire a traditional agency or consultant: sign a statement of work, they scope and deliver a project, then hand it back. Third, an embedded adoption pod: a contact inside your company (who knows where the work breaks) plus one outside AI specialist, taking one real workflow at a time and shipping it into production. The pod is the model I run. The full picture is in what an AI adoption pod is. The three differ less in what they can build than in how fast they reach a tool people actually use.
How do they compare on speed and cost?
Each path fails in its own way, and on a different clock.
- Hire in-house. Slowest to first result: in my experience, 3 to 6 months to recruit, onboard, and ramp before anything ships. And when your one AI hire leaves, the knowledge walks out with them.
- Agency or consultant. Faster, roughly 2 to 4 months from first call to production. The classic failure is shelfware: a polished tool delivered, demoed, and quietly abandoned once the consultants are gone. That's the stall MIT documented.
- Adoption pod. About 6 to 8 weeks to a first workflow in daily use, because a small autonomous team (Amazon's "two-pizza" logic) ships faster than a committee that plans forever. HBR calls that the experimentation trap.
On cost, in-house is a salary line whether or not tools ship; agencies bill per project; the pod is somewhere between, priced per workflow. The numbers are in what AI adoption costs and the cadence in how long it takes.
When does building in-house actually make sense?
Later than most people think. Building in-house is right when AI is core to your product, or when you already run several AI tools in production and maintaining them is a full-time job with steady work. At that point a permanent owner beats renting one. It's premature when nothing is live and you're hiring to figure out what to build: a generalist with no workflow to attach to produces demos, not adoption, and you pay a full salary through the slowest possible ramp. Build in-house to maintain and extend proven tools, not to discover your first one.
A simple rule to decide: under about 200 people with no ML team and nothing live yet, start with a pod and ship one workflow. Once you have three or four tools in daily use, or AI becomes core to the product, hire in-house to own them. Reach for an agency only for a one-off, bounded build you never intend to run yourself.
