A CTO’s guide to coding in the Age of AI

Yesterday, Joost Vunderink from Cloud++ stepped on stage with a Lego giveaway and a casual nod to his Commodore 64 roots, but quickly shifted gears into something every modern developer is either flirting with or overwhelmed by: AI code companions. His talk wasn’t your typical tech evangelism, it was a sharp, experience-rooted reflection on how AI is actually being used (and misused) in software development today.

AI code companions: helpful, not magical

The heart of the talk centered on what Joost calls the “AI code companion”: a context-aware assistant embedded in the developer’s IDE. Unlike ChatGPT in your browser, this AI sees all your code, knows your project context, and offers targeted support. It can autocomplete functions, suggest refactors, or even act as a tutor for junior devs.

But Joost was quick to remind us: it’s not magic. It’s a smart autocomplete. It doesn’t understand your business rules (yet). It won’t push back on bad architecture (yet). And it definitely can’t be trusted to make production decisions unsupervised.

He drew a distinction between four categories of AI:

  1. Automation – rule-based workflows (not really AI)
  2. Conversational AI – like ChatGPT
  3. Code companions – embedded, contextual IDE assistants
  4. Agents – semi-autonomous actors that execute tasks end-to-end

Cloud++ is all-in on #3 for now, experimenting with #4 (agents), but sticking to a human-in-the-loop philosophy.

The spaghetti problem (and other risks)

One of the most resonant moments came when Joost likened unfiltered AI-generated code to “kale and spaghetti.” It starts out healthy, structured, maybe even elegant. But feed the machine too much responsibility and suddenly your codebase is a steaming mess of entangled logic and half-baked assumptions. Developers become over-reliant, lose grip of system architecture, and mistakes creep in like mold behind the walls.

There were stories that landed hard:

  • AI inventing API calls that didn’t exist.
  • A lawyer using ChatGPT and citing made-up legal cases in court.
  • AI “solving” access-denied errors by removing authentication altogether.

The takeaway? “Trust, but verify” isn’t strong enough. You need processes, guardrails, and senior oversight.

The safe zone: what AI is actually great at

Here’s where Joost’s pragmatic approach shines. Cloud++ treats the AI like a smart intern:

  • Boilerplate generation (tests, scaffolding, documentation)
  • Refactoring (tidying up messy logic)
  • Explaining (onboarding new devs)
  • Setup (laying out initial project architecture)

None of these tasks involve mission-critical decisions. They’re grunt work, and AI excels at grunt work.

But even here, Cloud++ wraps everything in a policy. The AI gets a style guide. A coding philosophy. Explicit dos and don’ts. It’s coached like a junior developer. That’s how you avoid surprise spaghetti.

UX and UI with v0.dev: AI as creative muse

Joost then pivoted to design tools like v0.dev. Instead of writing Figma mockups from scratch, Cloud++ designers now prompt the AI for layout ideas, use the output for inspiration, then rebuild cleanly in Figma. It’s a kind of architectural sketchpad: fast, flexible, and often wrong, but useful.

Crucially, Joost made it clear: AI doesn’t replace design. It accelerates it. It’s the spark, not the fire.

Agents: fun side projects with serious potential

In a bonus segment, Joost shared how he built an AI agent that reads his email, checks his to-do list, and replies to messages, all through a self-hosted n8n setup. It was equal parts hobby project and glimpse into what’s coming: code reviewers that actually post feedback in GitHub, agents that regenerate documentation after each deploy.

The kicker? You still need a senior dev to check its work. Always.


Final takeaway: AI is your intern, not your architect

Joost’s message cuts through the AI hype: use these tools, but use them with eyes wide open. They’re powerful, but brittle. Fast, but sloppy. Confident, but not always correct.

The cleverness of Cloud++ isn’t in building the flashiest stack. It’s in how they train their AI like they train their juniors. They coach it. Shape it. Review its work. In a world of AI cowboys, that’s what makes their approach quietly revolutionary.

Become a member of the We Love SaaS community

Your best € 300 investment

we love saas community benelux