Karri Saarinen warns AI agents turn products into 'a collection of abilities' with no one behind them
Karri Saarinen argues that when each person ships changes through an agent, every change can make sense alone while the product's character averages out.
Saarinen writes that good software shows the fingerprints of its makers: what they chose to build, leave out, and keep polishing after it already worked. He says Linear runs on intuition formed by customer contact, not on process or data.
His worry is additive growth: features that work alone but weaken the whole. Agent-driven workflows could accelerate that.
Software with a point of view
In the post, Karri Saarinen argues that good software carries the experience, values and beliefs of the people who built it. Character comes from deliberate choices: what to build, what to leave out, and what to keep refining after it already functions.
He describes Linear's approach as valuing intuition over process or data, and defines intuition narrowly: "The intuition is your point of view on something." It is built from hands-on work, customer contact and mistakes. Before launch the team talked to people across companies and roles about how they worked.
Where agents come in
Many recent products grow by adding capabilities instead of rethinking core problems. His fear is that agent workflows extend this: when individuals use agents to make changes independently, each change may be sensible, but the combined result averages out distinctive choices and the product loses its guiding mind. His conclusion is optimistic: cheaper software creation should leave more room to refine a solution until its parts fit.
Builders agree the bottleneck is deciding, then split on whether a model can
jwr, who says they founded the parts-inventory SaaS PartsBox, described the same split from the maker's chair:
"It's in deciding what to build and what not to build, defining the overall character and vision of the product, looking at each screen and thinking "how does this help my customer do their job"."
They conceded that AI already out-codes them, and prmph, who says they lead a team building a data-analysis front end in Accra, pushed back on exactly that premise:
"a single session with Claude Opus is all it takes to debunk this, and yet people take it as given."
ttul, who says they founded MailChannels, supplied the pressure from the other side: companies told product teams to speed up releases, but
"the bottleneck isn't coding execution, it's choosing what to build and verifying that what got built doesn't suck."
He expects models to learn product management too. Whether or not that happens, the scarce input in every account is a person who owns the "no". The practical test for a team: can any single person say why a given feature is absent?
Get breakdowns like this every day
The top Hacker News stories, unpacked — by email, Telegram, or RSS.
Parser combinators in TypeScript, shown by parsing real ping output
Regex gets unreadable on nested formats and tree-sitter is heavy. A short blog series builds a typed middle path from plain functions.