I am Ben Monk. I build software, and I build the artificial intelligence features that run inside it.
Those sound like two jobs and in practice they are one. AI features live inside applications, and applications need designing, testing, deploying and maintaining. The interesting problems sit exactly where those two things meet, which is where I have ended up spending most of my time.
The work
Most of what I do falls into three areas.
Building software systems. The ordinary discipline of designing something, making it work, and keeping it workable while requirements change underneath it. Clear structure, tests that catch real problems, and code the next person can read.
Putting models into products. Retrieval, evaluation, guardrails, cost control, and the decisions about what a feature should do when the model gets it wrong. This is less glamorous than the demos suggest and it is where the actual difficulty lives.
Keeping it healthy afterwards. Observability, deployment, and the monitoring that catches a system quietly getting worse. AI features are unusually good at degrading without raising an error, so this matters more than it does elsewhere.
How I think about it
I have a strong preference for the smallest thing that solves the problem. Novel tools and elaborate architectures are costs paid later by whoever maintains the result, and that bill is easy to underestimate while the novelty still feels productive.
I care a lot about being able to tell whether something is working. A team that can measure whether a change helped makes better decisions than a team relying on impressions, and in work with probabilistic components that gap widens quickly.
I think the honest version of this field is less dramatic than the public conversation about it. The models are genuinely remarkable. Most of the work of making them useful is still careful engineering, and pretending otherwise produces things that demo well and disappoint in production.
Now
I am focused on applied AI engineering: retrieval systems, evaluation infrastructure, and the practical question of how to build features on top of components that will never be exactly right.
I write about that work in the blog, and the specific areas I work in are set out under expertise.
If you want to talk about a project, the contact page reaches me directly.