AI Lowers Build Costs but Raises Coordination Costs.

AI lowers the cost of building tools but raises the cost of coordinating them.

That was my takeaway from a recent HBR article, which focused on how work changed with AI. Their punchline was that AI didn’t shrink workloads. Instead, people simply started more projects with broader scopes and end up working more.

The article is linked in the comments, and the guidance it offers on how to prevent burnout is helpful.

However, reading between the lines, I noticed two other risks, and thought about how they might be addressed:

1️⃣ Blurred job scope ➡️ 𝗔𝘀𝘀𝗶𝗴𝗻 𝗮 𝘀𝗶𝗻𝗴𝗹𝗲 𝗮𝗽𝗽𝗿𝗼𝗽𝗿𝗶𝗮𝘁𝗲 𝗼𝘄𝗻𝗲𝗿

The article mentioned that product managers were writing code and engineers were spending more time coaching colleagues and checking work.

This potentially creates questionable accountability for the final product.

Also, inherited work creates its own risks. During my finance career, the times when I have taken over models have been more difficult than building models from scratch, as you must adjust to somebody else’s logic, instead of building something that’s intuitive to you.

Even with AI, the team needs to decide who is responsible for the final product and that person should be the one with the most appropriate expertise. If that’s the engineer, they may want to build themselves, instead of troubleshooting somebody else’s work.

2️⃣ Integration technical debt ➡️ 𝗗𝗲𝘀𝗶𝗴𝗻𝗮𝘁𝗲 𝗮𝗻 𝗶𝗻𝘁𝗲𝗴𝗿𝗮𝘁𝗶𝗼𝗻 𝗼𝘄𝗻𝗲𝗿

Less technical people building their own tools will likely multiply the siloed solutions that already exist in most organizations.

The challenge is that AI is currently better at creating tools than connecting them.

Recently, Claude helped me write an Excel macro in minutes, but I struggled when I tried to use Claude to help me build a Make dot com workflow. My experience is that AI is good with contained coding projects, but less reliable when asking for instructions on how to connect systems.

Every AI project should have an integration owner, even informally, to help ensure scalability and integration are considered early.

Everybody should experiment with AI, but not every project should survive, and one of the failure modes might be that the project is too siloed.

👉 I have been thinking more about how AI changes the way we work. What changes do you expect?