Engineering hygiene in the age of AI agents
AI agents are great and powerful. They are also a great magnifying lens on why we tried to establish engineering best practices and methodologies for the last 30+ years. Those are hard to ignore now as we intend to ship more, and at a higher rate. A rate that has never been seen before.
The tectonics of software development are radically changing. Even for an SMB with 20 engineers using AI agents moderately, the tingling feeling that comes and goes around PR review bottlenecks, release frequency, gatekeeping, testing, and platform stability is real.
We are literally sipping from a fire hose.
I regularly think about those organizations that were already not equipped to sustain a daily release frequency and are now using AI agents, believing their deficiencies no longer need fixing.
They traded a badly maintained rusty bike for the latest carbon fiber bike, but their cardio and VO₂ max are still the same and they have no clue how to improve. The muscle was never trained. If they ever were, now for sure neither the PowerPoint nor the weekly status meeting is sufficient anymore. Those engineering departments don’t have the fitness; they lack the conditioning. They keep throwing more AI tokens at the problem like a weekend warrior chugging protein shakes thinking it’ll make him stronger.
The answer is to get back to basics, with AI accelerating the setup.
Use AI to put a proper engineering foundation in place:
- Put a proper CI/CD in place. One you are not scared to run for deployments to production multiple times a day.
- Put a linter on everything.
- Write unit tests and end-to-end tests.
- Put a dependency management tool in place.
- Design API contracts and SDKs that allow your teams to integrate with each other.
- Start your feature with a design doc, or skip this phase and write ADRs (Architecture Decision Records) from the start.
- Derive C4 Model assets from them.
- Put proper telemetry in place. With OTEL there are no more excuses.
- Catch your exceptions and errors in production and review them regularly.
- Track your costs.
Focus your use of AI to set up all of the above. It will bring value, not the illusion of it. Before long, you’ll be able to ship faster because the fundamentals are in place, and they feed back into your coding agents.
The critical part is to not get complacent after the first few weeks by skipping the writing of an ADR, neglecting to update the C4 Model assets, or waving away an end-to-end test. Every single feature has to fit within this framework.
Those practices, once considered too time-consuming to implement or not worth the effort, can finally be adopted across an entire organization without tripling its headcount. They will feed your AI agents with the essential nutrients to build lean features and strengthen the connective tissue of your stack.
It will prevent you from medicating your engineering with more AI agents that are supposed to detect and fix what should never have been shipped in the first place.
Treat the cause, not the symptoms. Take care of your engineering hygiene so it works in symbiosis with your AI agents.
Ultimately, it serves customer satisfaction and happiness.
If you have suggestions or thoughts, keep the conversation going on my Substack.