Architecture
Trends in development platforms and what to do about them
AI coding agents are part of daily work, 90 percent of organisations have an internal platform, serverless has matured and Postgres is the default database.
Four developments are now big enough for a CTO to set a policy on, although none of them is new. AI coding agents are part of daily work, organisations build internal platforms, serverless has matured, and Postgres is the default database.
AI coding agents are just tools now
More than 30,000 people took part in the 2026 Stack Overflow Developer Survey, and Claude Code and GitHub Copilot are the most used coding agents in it. Of the developers who use AI coding assistants or agents, 73 percent use them daily. Google's 2025 DORA report shows the same picture: 90 percent of respondents use AI at work, and 30 percent report little or no trust in the code it produces.
I think that distrust is healthy. An agent writes code quickly, but someone still has to read it, test it and own it. Our developers work with tools like Claude Code every day, and that comes with rules: which tools you allow and what must never go into a prompt. How that works inside a team is covered in AI coding tools in a remote team.
My advice is to measure ticket lead time and the number of bugs after a release, not lines of code. That shows whether the tools help.
Platform engineering means fewer choices per team
According to the same DORA report, 90 percent of organisations have adopted at least one internal platform. The researchers see a direct correlation between the quality of that platform and how much value an organisation gets out of AI.
For a company with ten to thirty developers, a platform does not have to be a big project. It is a fixed path: one template repository, one CI pipeline, one way to deploy, secrets in one place and logging that works the same everywhere. New services and new developers follow that path, and so can an AI agent.
As a first step, turn your best-maintained service into a template repository, with the CI pipeline, the Dockerfile and the logging setup in it. The next service starts from that template, and every improvement goes back into the template instead of into a single project.
Give it an owner, even if that is only part of someone's week. Without an owner, local variations creep back in, and a year later you have five ways to deploy.
The hype around edge and serverless is over
Vercel itself now recommends migrating from the Edge runtime to Node.js, for performance and reliability. According to Vercel, runtime = 'edge' is no longer supported from Next.js 16.3, and routes and pages run on Node.js. On Vercel both runtimes run on Fluid compute, where you pay for active CPU time.
As long as your data lives in one place, a function close to your database helps you more than a function close to the user, and the gap between serverless and an ordinary server keeps shrinking. So do not rewrite anything for the edge, and pick a region close to your data. Aleksa works through that trade-off in serverless or your own servers.
Postgres is the default
In the 2026 survey, 57.9 percent of respondents had used PostgreSQL in the past year. That makes it the most used database, well ahead of SQLite (42.1 percent) and MySQL (35.1 percent). The large data platforms are buying in as well: in May 2025 Databricks announced it would acquire Neon, a serverless Postgres company, and in June 2025 Snowflake announced its acquisition of Crunchy Data.
For a new product I would not pick anything else without a good reason. Aleksa describes what Postgres covers and where its limits are in Postgres as the default database.
What I look for in new developers
When I hire a developer now, I look at two things on top of the usual skills. Can they judge AI output and throw it away when it is wrong? And do they know how their own code gets to production? A stack like TypeScript, Node.js and Postgres fits that well, because a lot has been written about it and the tooling is mature.
What I would not do is chase every trend. A boring default stack with good tests gets more out of new tools than a stack that changes every year. The stacks we use are listed on our technology page.
Where I stay cautious
I do not know whether agents will deliver complete features on their own a year from now, and I would not build a plan on it. What I do see is that a team with good tests and one fixed path to production gets more out of every new tool. The DORA report reaches the same conclusion: the greatest return comes from the quality of internal platforms, clear workflows and teams that are aligned.
Sources
Frequently asked questions
Which AI coding agents do developers use most?
In the 2026 Stack Overflow Developer Survey, Claude Code and GitHub Copilot are the most used coding agents. Of the developers who use AI coding assistants or agents, 73 percent use them daily.
What is platform engineering?
A fixed, shared path for teams to build, test and ship code, with templates, pipelines and tooling that have one owner. According to the 2025 DORA report, 90 percent of organisations have adopted at least one internal platform.
Is PostgreSQL the most used database?
Yes. In the 2026 Stack Overflow Developer Survey, 57.9 percent of respondents used PostgreSQL in the past year, ahead of SQLite (42.1 percent) and MySQL (35.1 percent).
About the author
Bas Leijder Havenstroom
Managing director
Together with Erik Schoorl, Bas founded ESBL, one of the two companies behind Brainpower Group. He is managing director of ESBL B.V. and the first point of contact for clients.
More from Bas
