A few days in the Netherlands visiting Dave Ebbelaar— and of course we tested the new Jev model from typsafe: 30 seconds to classify 500 emails.
We had a really good discussion about what it actually takes to build AI products and services that genuinely get adopted inside companies.
One of the biggest conclusions: don’t try to scale too early.
First, you need to do things almost manually with your first customers. Sit next to them, see where they get stuck, understand what employees are actually asking for, and figure out which problems are truly worth solving.
Only then do you turn those lessons into processes, skills, agents, and automation.
A few ideas that stayed with me:
AI in companies is not won through technology. It is won through adoption.
👉 Your own company is the best demo.
If you can show how you already use AI internally and what results it produces, the story becomes much more credible than any theoretical presentation. That’s exactly what we’re doing at Termene.ro with our fleet of agents.
👉 AI should be introduced bottom-up, not just top-down.
Instead of management deciding every AI project, the people actually doing the work should be able to ask agents for help directly — in Slack, Teams, or the tools they already use. That’s where some of the best use cases emerge.
👉 Don’t automate a bad process.
Sometimes the real problem is not the six-hour task you can see on the surface, but the entire workflow behind it. AI forces us to rethink the process, not just make it faster.
👉 The first implementations need to be very hands-on.
The first 5–10 companies may even be unprofitable or difficult to serve. But that’s exactly where you discover what is repeatable, what needs to be built, and what can later be automated.
👉 One interesting direction is to make agencies, consultants, and forward-deployed engineers the implementation channel, using a common platform (@sferal.ai) instead of rebuilding the infrastructure from scratch every time. A training academy could also accelerate the development of this ecosystem.
And perhaps the biggest opportunity of all:
Those of us who work with AI agents every day can easily feel like the market is already very advanced.
In reality, the gap between the technological frontier and adoption inside normal companies is still enormous.
That’s exactly why onboarding, education, and implementation are just as important as the product itself.
My takeaway from the discussion:
Service first. Learn fast. Productize what repeats. Automate only after you understand the customer.
That may be how the next wave of AI companies gets built. Sferal AI Termene.ro
We had a really good discussion about what it actually takes to build AI products and services that genuinely get adopted inside companies.
One of the biggest conclusions: don’t try to scale too early.
First, you need to do things almost manually with your first customers. Sit next to them, see where they get stuck, understand what employees are actually asking for, and figure out which problems are truly worth solving.
Only then do you turn those lessons into processes, skills, agents, and automation.
A few ideas that stayed with me:
AI in companies is not won through technology. It is won through adoption.
👉 Your own company is the best demo.
If you can show how you already use AI internally and what results it produces, the story becomes much more credible than any theoretical presentation. That’s exactly what we’re doing at Termene.ro with our fleet of agents.
👉 AI should be introduced bottom-up, not just top-down.
Instead of management deciding every AI project, the people actually doing the work should be able to ask agents for help directly — in Slack, Teams, or the tools they already use. That’s where some of the best use cases emerge.
👉 Don’t automate a bad process.
Sometimes the real problem is not the six-hour task you can see on the surface, but the entire workflow behind it. AI forces us to rethink the process, not just make it faster.
👉 The first implementations need to be very hands-on.
The first 5–10 companies may even be unprofitable or difficult to serve. But that’s exactly where you discover what is repeatable, what needs to be built, and what can later be automated.
👉 One interesting direction is to make agencies, consultants, and forward-deployed engineers the implementation channel, using a common platform (@sferal.ai) instead of rebuilding the infrastructure from scratch every time. A training academy could also accelerate the development of this ecosystem.
And perhaps the biggest opportunity of all:
Those of us who work with AI agents every day can easily feel like the market is already very advanced.
In reality, the gap between the technological frontier and adoption inside normal companies is still enormous.
That’s exactly why onboarding, education, and implementation are just as important as the product itself.
My takeaway from the discussion:
Service first. Learn fast. Productize what repeats. Automate only after you understand the customer.
That may be how the next wave of AI companies gets built. Sferal AI Termene.ro
41