Insights

Insights

Observations on AI systems, product strategy, and the operational patterns that separate working infrastructure from abandoned experiments.

Aerial view of a road forking through forest
4 min read

Should You Build AI In-House? An Honest Answer.

When to build your AI in-house, when to bring in a partner, and the one test that decides it either way, from someone who sells the second option and will still tell you to build.

38shiftRead
An abstract web of connected, glowing network nodes
7 min read

We Shipped WebMCP. Here Is What That Means.

Chrome's new agent-tools API, the implementation we put on 38shift, and the strategic case for instrumenting the agent-readable web first.

38shiftRead
Long-exposure light trails curving through darkness
7 min read

Inside a Signal-Driven Growth Engine

A field report on GTM Engineering: signals, agents, and why outbound stopped being a headcount problem.

38shiftRead
Symmetric bundles of data-center cabling converging
6 min read

How 38shift Runs on AI Agents: Our Internal Stack

What we learned by automating core internal operations with the same kinds of systems we build for clients.

38shiftRead
A geometric diamond-patterned building facade
4 min read

Inside the Digital Maturity Index of Kosovo's Banking Sector

A summary of the Digital Maturity Index benchmark, which scores Kosovo's commercial banks across digital and AI capability and groups the sector into three clear tiers.

38shiftRead
Minimal concrete architecture lit from above
2 min read

The Governance Layer Every AI Product Needs

As soon as AI moves beyond experimentation, governance stops being a legal afterthought and becomes part of the product itself.

38shiftRead
An interlocking steel structural framework
2 min read

Building AI Systems Companies Can Run

Most AI initiatives fail at the transition from demo to daily operations. The difference is almost never model quality alone.

38shiftRead
A dense tangle of overhead wires against the sky
2 min read

The Silent Killer of Velocity: Over-Engineering for Scale on Day One

Startups burn through runway by building Kubernetes clusters for 10 users. Here is a pragmatic approach to scaling tech infrastructure.

38shiftRead

Have a workflow worth automating?

Tell us where the manual work is and we will scope what we would build.

Start a project