LLenny's Podcast
← All episodes
Dan Shipper24 May 2026

The AI paradox: More automation, more humans, more work

6Frameworks
14Insights

Frameworks in this episode

Innovation3 steps

Build Your Own Senior-Engineer Benchmark

Measure AI honestly by scoring new models against real human experts rewriting your actual broken work.

Read →
Mindset3 steps

Frozen Competence: Where Human Value Survives

Models commoditize yesterday's competence; your value is using that cheap competence to make something new.

Read →
Self-Mastery3 steps

Ride the Models

Stay valuable in AI by playfully applying each new model to your own work and re-testing what it couldn't do before.

Read →
Leadership3 steps

Super Agent With a Forward-Deployed Human

Give the whole company one shared agent, owned by a human who keeps it working — not personal agents for everyone.

Read →
Communication3 steps

The AI-Document Slop Test

An AI-written document is fine — as long as you can stand behind every line and it took longer to make than to read.

Read →
Productivity3 steps

The Reach Test

Judge whether an AI tool is truly useful by whether you reach for it unprompted each morning.

Read →

Insights & moments

The myth-busts, hot takes, explainers, and tools worth keeping.

Hot Take· 7

Hot Take13:30

Why He Flipped to One Super Agent Per Company

Shipper was originally convinced everyone would have their own personal agent, a parallel shadow org chart. He completely flipped. Personal agent harnesses like Open Claw turned out to be too much work and constantly break, and an agent is only useful when a human cares about it and watches it. So companies like Shopify and Ramp are converging on one super agent maintained by a forward-deployed engineer.

  • He originally believed in a personal agent per person, then completely reversed
  • Open Claw-style personal agents break constantly and demand too much maintenance
  • An agent needs a human who cares about it, or it stops being useful
  • The current working model is one company-wide agent maintained by a forward-deployed engineer
  • Personal agents will return once models get independent enough to not need fiddling

“And I have completely flipped.”

Dan Shipper · 13:30

“in order for an AI agent to be useful right now, it really needs a human who cares about it.”

Dan Shipper · 14:30
#agents#org-design#future-of-work
Hot Take31:00

CLIs Are Over

When Claude Code got popular, many people concluded the terminal was the future and moved to working from the CLI. Shipper disagrees: the CLI was not the reason Claude Code worked. Once you move into a real GUI you get the same benefits with a nicer experience, and most of his technical team no longer uses the CLI as their main work surface.

  • People wrongly credited Claude Code's success to it being a CLI
  • GUIs exist for a reason and deliver the same benefits more pleasantly
  • Most technical people at Every no longer use CLIs as their main surface
  • CLIs won't disappear entirely, but the CLI era peak has passed

“CLIs are over. Um we we speed ran the CLI uh era. It was nice while it lasted, but I think”

Dan Shipper · 31:00
#cli#developer-tools#hot-take
Hot Take37:00

Buy SaaS Stocks: The SaaS Apocalypse Is Done

In one of his most contrarian calls, Shipper says he would buy SaaS stocks right now and expects them to rise majorly. His reasoning: agents don't kill SaaS, they multiply its users. Every internally has agents everywhere yet their SaaS spend is up year over year, and agents will hit these products at very high volume.

  • He would buy SaaS stocks and expects big gains over the next couple of years
  • Agents increase the number of SaaS users rather than replacing SaaS
  • Every's own SaaS spend is up year over year despite heavy agent use
  • Agents will drive an insane spike in demand and volume for SaaS products
  • Delivered explicitly as 'not investment advice'

“I would buy SAS stocks right now.”

Dan Shipper · 37:00

“what agents do is increase the number of users of SAS. Not get rid of it.”

Dan Shipper · 38:00
#saas#investing#hot-take#contrarian
Hot Take39:00

Automation Is a Lie: Every Agent Needs a Human

Every doubled its headcount in a year despite being an extremely AI-forward company. Shipper's explanation is that automation is a lie: every time you automate something you need a human on top making sure it works. He compares working with AI to being a manager, and points out that managers actually spend a lot of time working, not relaxing on a beach.

  • Every doubled in size while being aggressively AI-forward
  • Every automation still needs a human ensuring it works well
  • Working with AI resembles being a manager (the 'allocation economy')
  • Managers spend a lot of time working, not sitting idle

“Automation is a lie.”

Dan Shipper · 39:00

“managers actually spend a lot of time working. Most managers are not like on the beach.”

Dan Shipper · 39:30
#automation#hiring#management#hot-take
Hot Take1:08:30

Super Bullish on PMs: The Marcus Story

Shipper is extremely bullish on product managers who get AI-pilled. His anecdote is Marcus, a PM by training who ran Axios's writing product, took a year off to get deep on AI tools, and now runs Every's writing app Spiral. Only lightly technical, Marcus pairs limited coding knowledge with spiky product sense and now ships faster than almost anyone on the team.

  • PMs who get deeply AI-pilled are a major winner in Shipper's view
  • Marcus is a PM by training who previously ran Axios's writing product
  • He is only lightly technical but knows things like a database migration
  • He pairs that with spiky product sense and ships faster than almost anyone
  • Coding models have gotten good enough to make this pairing possible

“I am super super bullish on PMs.”

Dan Shipper · 1:08:30

“it makes me very, very bullish on any PM who gets like really AI pilled.”

Dan Shipper · 1:10:00
#product-management#careers#hot-take
Hot Take1:11:00

Full-Stack Designers Are a Superpower

The other group Shipper thinks will be super-powered is full-stack designers. Designers who live in these tools feel newly empowered to build the beautiful interactions engineers used to resist or botch. Because most AI-built UI looks like generic slop, designers who can make things look genuinely different, and now actually ship them via pull requests, stand out and have a huge opportunity to start their own things.

  • Designers can now build the interactions engineers used to refuse or fumble
  • Most vibe-coded UI looks the same, so distinctive design stands out from slop
  • Designers increasingly ship pull requests directly instead of handing off
  • There's a big opportunity for these designers to start their own companies

“they can make stuff that looks so different and now they can actually build it.”

Dan Shipper · 1:11:30
#design#careers#hot-take
Hot Take1:18:30

The Edge of AI Isn't in San Francisco

Shipper pushes back on the idea that the frontier of AI lives in San Francisco. The people there make the models but don't actually know all the ways to use them. The real edge is wherever AI meets a real human doing something specific, so anyone who consistently applies new models to their own work gets to be among the first to discover what those models are useful for.

  • The edge of AI is not where the models are built
  • SF builders make the models but don't know all the ways to use them
  • The frontier is wherever AI meets a real human doing real work
  • Every, based in Brooklyn, feels ahead of SF because they use models for everything

“I think people think of the edge of AI as being in San Francisco. And I actually don't think that that's where it is.”

Dan Shipper · 1:18:30

“I think the edge of AI is wherever AI meets like a real human doing something.”

Dan Shipper · 1:18:30
#ai-adoption#hot-take#future-of-work

Explainer· 3

Explainer11:30

Work Will Split Two Ways: A Super Agent and an Agentic OS

Dan Shipper predicts how work changes in the coming year will bifurcate into two modes. First, everyone in a company will have at least one agent they can talk to and offload work to. Second, most work you do will actually happen on your computer inside an environment like Codex or Claude Co-work, which becomes the operating system for your email, documents, and everything else.

  • Mode one: everyone gets at least one agent they delegate work to, probably in Slack
  • Mode two: your day-to-day work moves into Codex or Claude Co-work as a work surface
  • That surface becomes the effective operating system for email, docs, and more
  • This is described as the emerging competitive landscape, not a fixed timeline

“it's going to bifurcate in this in two main ways, how you how you use agents.”

Dan Shipper · 11:30

“most of the work that you do is actually going to happen on your computer in an environment like Codex or Cloud Co-work”

Dan Shipper · 12:00
#future-of-work#agents#codex#claude-code
Explainer23:30

SaaS Will Run Inside Your Agent, and Users Bring Their Own Tokens

Instead of AI being baked into SaaS tools, Shipper predicts SaaS tools will run inside Codex or Claude Code via an in-app browser. When you run your agent against a website, you spend your own tokens, not the vendor's. This changes SaaS margins because the company no longer pays for the AI, and it changes what you build: simpler products designed for humans and agents to collaborate on together.

  • SaaS apps get used through the agent's in-app browser, with your computer's access
  • Users bring their own tokens, so the SaaS vendor stops paying for AI usage
  • This restores SaaS margins rather than eroding them
  • Products get simpler because the agent handles formatting and grunt work
  • You now design software for humans and agents to work on the same thing at once

“When I run the agent on that website, I'm using my tokens. I'm not using the the vendor's tokens.”

Dan Shipper · 24:00

“you build it now for both humans and agents to use at the same time and it changes your margins”

Dan Shipper · 25:00
#saas#business-model#agents#codex
Explainer1:13:30

Models Make Yesterday's Human Competence Cheap

Shipper argues the AI job apocalypse isn't real, and explains why structurally. What models do is ingest what has already happened and make yesterday's human competence cheap and instantly deployable. That competence gets commoditized because everyone uses the same models in the same default way, which leaves room for humans to take that frozen competence and make something new and interesting.

  • Models make yesterday's human competence cheap and easy to deploy anywhere
  • Because everyone uses the same models the same way, output looks identical
  • That sameness commoditizes the competence, so it stops being valuable
  • Humans create value by taking the frozen competence and making something new
  • Model incentives to be aligned and compliant keep them trailing frontier human work

“what models do in general is they make yesterday's human competence cheap.”

Dan Shipper · 1:13:30

“it becomes commoditized. Like it's not valuable anymore.”

Dan Shipper · 1:14:30
#ai-economics#jobs#future-of-work

Story· 1

Story40:30

The Senior Engineer Benchmark and Vibe Coder Elbow

Shipper vibe-coded his app Proof on the side, and after launch the servers kept crashing every ten minutes while Codex couldn't fix it. He vibe-coded so hard he got bursitis in his elbow, then hired two senior engineers to rewrite it independently, creating a benchmark. Models scored ~30/100 versus a human's high-80s to low-90s, until GPT 5.5 jumped to 62, illustrating both real progress and how far AI still has to go.

  • Proof was fully vibe-coded and crashed roughly every ten minutes after launch
  • Codex kept fixing one bug and causing four more, going in circles
  • He got bursitis in his elbow from vibe coding so intensely
  • Two human senior engineers rewrote the codebase to form the benchmark
  • Models scored ~30/100 vs a human's high 80s to low 90s; GPT 5.5 hit 62

“I I I vibe coded so hard I got bursitis on my elbow.”

Dan Shipper · 41:30

“And all the models until GPT 5.5 got like a 30 out of 100. And senior like a human senior engineer gets like high 80s,…”

Dan Shipper · 42:00
#benchmarks#vibe-coding#codex#story

Takeaway· 3

Takeaway33:30

Two Agents Are Better Than One

Shipper is not an agent maximalist and expects people to use many agents. A key insight: when your agent talks to another company's agent, it can convey far more context about you and what you want than you could type yourself. Building products that assume users arrive through Codex or Claude Code lets agents onboard and negotiate on your behalf.

  • Expect many agents rather than one all-powerful agent
  • Your agent can pass rich context to another agent faster than you can type it
  • Products that assume agent-mediated users can skip manual onboarding flows
  • Agents talking to agents creates a custom, high-context experience

“And I really do think that two agents are better than one.”

Dan Shipper · 33:30

“When I have Codex interact with another agent, it can give so much more context about me and what I want than I would be…”

Dan Shipper · 33:30
#agents#product-design#future-of-work
Takeaway1:16:00

To Keep Your Job, Ride the Models

Asked what to do to avoid being laid off, Shipper's answer is simple: ride the models. Use them for whatever you do, and when new models drop, try them and figure out how their new powers apply to your work instead of ignoring them out of fear. The practical method is to be curious and playful and keep 'turning over rocks' to test what a new model can now do.

  • The one thing to do is ride the models: use them for whatever you do
  • When new models ship, try them and find their new powers rather than ignore them
  • Fear-based avoidance is understandable but leaves you behind
  • The method is curiosity and play, retesting tasks a model couldn't do before

“the only thing you need to do is ride the models. And that means use them for whatever it is that you do.”

Dan Shipper · 1:16:30
#careers#ai-adoption#takeaway
Takeaway1:24:00

Find Your Moment of Joy With AI

Shipper's closing advice is to approach AI through play rather than FOMO. The best way to find genuinely useful things to do with AI is to do something enjoyable with it. He cites Nikhil Singhal's framing that you have to find your moment of joy, the moment you can't believe AI just did something for you, which is what keeps you building.

  • Too many people engage with AI out of fear of missing out or losing their job
  • The best way to find useful uses is to do something enjoyable with it
  • Nikhil Singhal's framing: find your moment of joy with AI
  • That first 'I can't believe it did this' moment is what sustains building

“you got to find your moment of joy with AI.”

Dan Shipper · 1:24:30
#ai-adoption#mindset#takeaway