Working with AI

We jumped on the AI train at my current company, and have all been fully riding that train since about February 2026. We code with Claude and we have a review bot to review our changes. It has accelerated our speed. It has overwhelmed us. It has allowed us to ship features and core functionality we had been dreaming of for years. It has left us a little less aware of what is actually happening in our system. It has given us the power to resolve customer bugs in record time. It has allowed us to merge not-so-performant code that takes half the system down. It has allowed us to ingest and support so many more sources. It has allowed us to generate code that is plain and simple wrong. It has inspired us to create at a speed we've never been able to achieve before. It makes it easy for someone to merge not-so-great code and features into the system with little oversight.

The struggle is real when working with AI. As the person who is usually writing up issues and features for others, it has been a game-changer for me - I can suddenly complete so many more changes. My brain is also completely dead at the end of the day; I've used up all my processing power about 6-7 hours into my day. I've also watched some colleagues and employees completely lose their sense of structure as they use more AI in their day-to-day. The standup status that just lists the claude tabs the engineer has open. The response that is clearly directly from claude, not interpreted by an engineer. The general overwhelm. The drop in motivation that comes with the overwhelm. The just-below-the-surface frustration that lives with being overwhelmed.

How we handle the overwhelm? We're working on it.

• Less large topics per person.
• Tracking for everything so it doesn't get lost.
• Automation everywhere we can have it.
• Monitoring, so much monitoring.
• And alerts, too - since that is what makes the monitoring actionable.

The problems AI has presented us with as a team helped us mature into an org that truly considers and understands the performance implications of their features, and thinks about features in a performance-first way. We still have a ways to go, more guardrails and automation to make it all feel a little safer and saner, but it's nice to see that the chaos introduced allowed us to mature, not deteriorate, as an engineering org.