Overview of The AI paradox: More automation, more humans, more work | Dan Shipper
In this episode, Lenny Rachitsky talks with Dan Shipper, CEO of Every, about how AI is actually changing work inside a highly AI-forward company. Dan’s core thesis is contrarian: AI will not cause a broad “jobpocalypse.” Instead, it will create more work, more software, more review, and more need for humans. He argues that the future of work will split into two main surfaces: a company-wide agent you delegate to in tools like Slack, and an AI-native work environment on your computer in products like Codex or Claude Code. He’s also bullish on SaaS, PMs, and full-stack designers, and skeptical of the idea that CLIs or standalone agent hype will dominate forever.
Key Predictions About the Future of Work
1) Every company will have a “super agent”
Dan predicts that the first wave of agent adoption will be one shared agent per company, not one personal agent per employee.
- This agent will likely live in Slack or a similar work communication tool.
- It will handle async tasks like data requests, research, coordination, and routine ops.
- Over time, more specialized team agents may emerge underneath it.
His key insight: agents need humans who care about them. If no one maintains, reviews, and corrects them, they quickly become useless.
2) Most work will move into AI-native work surfaces
The bigger shift, in Dan’s view, is that much of knowledge work will happen inside tools like Codex, Claude Code, or Cowork.
- These tools become the new operating system for work.
- Instead of AI being embedded into SaaS, SaaS tools will be used inside the AI environment.
- Humans and agents will work side by side on the same task, with shared visibility.
This is a major UX and infrastructure shift, not just a feature shift.
3) CLIs are a temporary phase
Dan believes the recent “CLI era” was speedrun and will not be the final form.
- CLIs were useful for agents because they exposed more power.
- But for most users, GUIs are still better, especially for non-technical work.
- The future is not “everything becomes terminal”; it’s “agents and humans collaborate in a UI.”
How the Shape of Work Will Change
More people will do technical work
AI makes it easier for non-technical people to create pull requests, edit products, and contribute directly to software.
- PMs, editors, ops folks, and designers will increasingly ship changes themselves.
- Technical teams will spend more time reviewing, integrating, and quality-controlling that output.
- Work becomes less about doing everything manually and more about coordinating and curating output.
More review, more coordination, more system design
As AI increases output volume, the bottleneck moves downstream.
- More code means more review.
- More analysis means more checking.
- More documents and emails means more triage.
- More capability means more need for coherent system design.
Dan repeatedly emphasizes that automation doesn’t eliminate oversight—it creates a new layer of human work on top.
New role: forward-deployed AI engineer
Dan thinks a new category of role is becoming real: the forward-deployed engineer for AI systems.
These people:
- Set up and maintain company agents
- Make sure the agents work reliably
- Help non-technical teams use AI safely and effectively
- Debug the messy edge cases between humans, software, and agents
In other words: automation creates the need for people to run the automation.
AI-generated writing becomes normal internally
Dan argues that people will get more comfortable with AI-generated documents and emails, especially for internal work.
- Strategy docs, planning docs, and long emails are often better when AI assists.
- What matters is whether the human stands behind the content.
- “Slop” is the problem, not AI writing itself.
He expects AI-assisted writing to become standard for internal and operational work.
Who Will Be Most Successful
PMs
Dan is extremely bullish on product managers who get AI-native.
Why?
- AI reduces the need to organize large teams for every task.
- PMs can now prototype, ship, and iterate much faster.
- Strong product sense becomes even more valuable when building is easier.
Full-stack designers
Designers who can build directly will thrive.
- They can turn ideas into working products without waiting on engineering.
- Their creative edge helps them avoid the generic look of AI-generated slop.
- Dan thinks this may open entrepreneurship opportunities for designers.
Generalists and people who “ride the models”
Dan’s favorite phrase is essentially: ride the models.
The winners will be people who:
- Try new models as soon as they come out
- Play with them on real problems
- Keep asking, “Can this do more now than it could before?”
- Stay curious and adaptive instead of fearful
Less changed roles: sales, some leadership functions
Dan suggests that some roles, like sales, may be less fundamentally changed than engineering or design.
- Top-of-funnel, research, and qualification get easier.
- But the core work of persuasion and relationship-building remains human.
- CEOs and senior leaders may look “unchanged” for now, but he thinks they actually need to be deeply AI hands-on.
Bigger Business Takeaways
SaaS is not dead
Dan strongly rejects the idea of a “SaaSpocalypse.”
His view:
- Agents will increase SaaS usage, not destroy it.
- Users will bring their own AI/token usage into products.
- SaaS companies should optimize for human + agent collaboration.
- This may even improve margins and accelerate demand.
Build for humans and agents together
Future software should assume:
- The human is using the web UI
- The agent is using the CLI or backend
- Both need visibility into the same work
- Approvals, logs, rollbacks, and summaries become essential
This is a different product paradigm than building just for humans.
Recommended Actions for Listeners
If you want to stay relevant in this future, Dan recommends:
- Use Codex / Claude Code / similar tools daily
- Try your real workflows inside AI-native environments
- Learn how to work with agents, not just around them
- Make your products agent-friendly if you build SaaS
- Experiment with a company Slack agent or internal bot
- Find a problem you actually care about and solve it with AI
- Play, don’t just doomscroll the hype
His strongest practical advice: don’t be afraid of the models—ride them.
Notable Quotes and Ideas
- “Automation is a lie.”
- “Every agent needs a human.”
- “The AI jobpocalypse is not really a thing.”
- “Models make yesterday’s human competence cheap.”
- “Creativity is going to be more and more valuable.”
- “CLIs are over. We speed-ran the CLI era.”
- “The SaaSpocalypse is dumb.”
Final Takeaway
Dan’s worldview is paradoxical but coherent: AI will automate more than ever, yet humans will remain more necessary than people expect. The real winners will not be those who resist AI, but those who use it deeply, adapt quickly, and build systems where humans and agents work together.