AI makes building cheap, not owning.

Video caption

AI makes building cheap, not owning.

Each local win creates support outside the maker's job.

Owner, users, failure cost, maintenance budget and exit path.

What I would require: Fund custom advantage; buy commodity reliability.

#EricFieldNotes

Full written post / accessible read

When everyone has Claude Code or another capable agent, every recurring annoyance looks buildable. That is useful power. It also tempts a company to create a fleet of tiny products without product managers, support plans or a reason to compete with maintained SaaS.

Picture ten teams each replacing a small paid service. Each tool works for its creator. Then permissions, exports, handoffs and exceptions arrive. People spend their attention asking agents for repairs while their original work waits. The debt appears as fragmented focus, not a single failed project.

For each proposed internal build, document who will own it, who relies on it, what happens on failure, what acceptance tests exist and how data leaves it. Compare the total with buying a maintained tool. Use a time-boxed trial only when the expected advantage is specific.

Let agents implement the bulk of genuinely strategic software, and hold engineers responsible for architecture, acceptance and operations. For commodity functions, pay the small fee when the maintained service wins the full comparison. Do this because a business cannot outsource judgment to a pile of cheap prototypes.

#EricFieldNotes

Evidence and boundary

On-screen label: ILLUSTRATIVE BUILD/BUY. Illustrative cases are not measured incidents. Research papers and vendor documents support the stated mechanism only within their studied or documented scope.

• METR randomized trial (S20)
• METR 2026 productivity experiment update (S93)