KnowledgeSketches|← All EpisodesLenny’s PodcastResearch papers
From Lenny's Podcast
Knowledge Sketch

The AI Paradox:
More Automation, More Humans, More Work

Dan Shipper
CEO and Founder, Every
MAY 24 2026
The Paradox

More AI,
More Human Work

MORE AIMORE WORK
"We have so much automation, so much AI, and I also work way more."
  • "The AI job apocalypse is not really a thing." His feeling is that in a year humans will have more work to do, even as models get better
  • Every agent needs a human who cares about it. The minute you sever that connection, the agent "is not really that useful anymore"
  • Models make yesterday's human competence cheap, so it gets commoditized and is not valuable anymore
  • The edge of AI is wherever it meets a real human doing something, not San Francisco
Framework

Two Surfaces for All Your Work

SUPER AGENT IN SLACKCODEX / CLAUDE COWORK
Dan's predictionHow you work with agents is "going to bifurcate in two main ways." One is an agent you delegate to. The other is a work surface on your computer where most of your work happens.
  • The super agent: for now, one agent for the whole company that people offload work to, probably in Slack
  • The work surface: an environment like Codex or Claude Cowork becomes an operating system for all your work, your email and your documents included
  • SaaS inside the agent: tools run in the agent's in-app browser, and the user brings their own tokens, not the vendor's
  • Two agents are better than one: your agent talks to the app's agent and gives it more context about you than you would be able to type
Playbook

Automation Still Runs on People

"In order for an AI agent to be useful right now, it really needs a human who cares about it."
  • "Automation is a lie." Every time you automate something, a human sits on top making sure it works
  • The forward deployed engineer is "for real" as a new job: someone responsible for making sure the agent is working for the whole company
  • Benchmarks make AI look more autonomous than it is. They rise on problems people have already framed and can score
  • Suddenly everyone is an engineer, "but that doesn't mean we fire the engineers. There's way more demand for engineers"
30
every model before GPT 5.5
62
GPT 5.5
~90
a human senior engineer
Dan's senior engineer benchmark, out of 100. His numbers are rough: "like a 30", "like a 62" (he also calls it 60), and "high 80s, low 90s" for a human. GPT 5.5 got its score using an Opus 4.7 plan.
The senior engineer test

Dan vibe coded an app called Proof, and it kept going down after launch. Two senior engineers each rewrote it. Now he asks each new model to rewrite it from first principles. Other models do a little patch. GPT 5.5 was the only one that ripped out old code.

Why humans stay

The thinking it takes to frame a problem and write it down cannot be measured. So even if the benchmarks get saturated, that does not replace all senior engineers. "I still hire engineers."

Playbook

Ride the Models

  • Ride the models: use them for whatever it is that you do, and try each new one when it comes out
  • When a new model comes out, turn the rock over again. "It may not work now, but it may work eventually"
  • Have fun with it, not FOMO. "The best way to actually figure out interesting, useful things to do with AI is to do something enjoyable"
  • "I am super, super bullish on PMs and full stack designers." His example is Marcus, a "lightly technical" PM who ships faster than almost anyone on the team
Where the edge is"The edge of AI is wherever AI meets a real human doing something." Whenever a new model comes out, you get to be one of the first people to discover what it might be useful for.
Contrarian

What Everyone Gets Wrong

✗The AI job apocalypse is comingINSTEAD →✓ "The AI job apocalypse is not really a thing." He does not think the mass unemployment some AI CEOs talk about will happen.
✗AI kills SaaS, the SaaSpocalypseINSTEAD →✓ "I would buy SaaS stocks right now." What agents do is "increase the number of users of SaaS, not get rid of it." He adds: "Not investment advice."
✗Everyone is moving to the CLI and the terminalINSTEAD →✓ "CLIs are over. We speed ran the CLI era." He adds that CLIs will not go away completely, but "it's just nicer to be in a GUI."
✗More automation means less human workINSTEAD →✓ "Automation is a lie." Every time you automate something, you need a human on top of it. Dan says he works way more.
Based on Dan Shipper's episode on Lenny's Podcast. All ideas on this page are from the episode.Read on Lenny's NewsletterFollow @danshipper on X
Every sketch here began as a transcript.

Send us yours and we’ll draw it. $1 a sketch, or $49 for all 11 formats.

Sketch mine →
𝕏︎ X / Twitterin LinkedIn📸 Instagram🔗 Copy link
0:00