#554. AI悖论：自动化越多，人越重要，Dan Shipper 预测未来一年工作方式巨变

Episode #554. AI Paradox: The More Automation, the More Humans Matter - Dan Shipper Predicts Major Changes in Work Styles Over the Next Year

📝 Episode Summary

In this episode, we cloned: The AI Paradox: More Automation, More Humans, More Work | Dan Shipper from the top tech and startup podcast *Lenny's Podcast* **The AI paradox: More automation, more humans, more work | Dan Shipper**

Our guest Dan Shipper is the CEO and founder of Every, one of the few entrepreneurs who deeply integrate AI into their company's daily operations. A year ago, he predicted that Claude Code would be underestimated, especially in non-engineering contexts. This prediction has proven to be highly accurate. Therefore, Lenny invited Dan back to the show to discuss his bold predictions about AI work styles for the next year in detail.

The core idea of this episode is a counterintuitive "AI Paradox": as AI automation increases, human work may not decrease but rather increase, with more jobs requiring human judgment, management, integration, and creativity. Dan believes that future knowledge work will primarily occur in two types of interfaces: one is a super Agent within companies, possibly accessible via Slack by everyone; the other is AI work operating systems like Codex, Claude Code, and Cowork, where users will incorporate browsers, documents, emails, and SaaS tools into these environments.

Dan also presents many counter-consensus views: SaaS will not die but could become stronger due to Agent usage; each Agent needs a person; PMs and full-stack designers will become more valuable in the AI era; AI will not simply lead to job apocalypses, but those who do not "follow the models" will be left behind. This episode is a high-density discussion on the future of AI work, organizational design, SaaS business models, and personal career strategies.

👨‍⚕️ Guest Introduction

Dan Shipper, CEO and founder of Every. Every is a media and product company focused on AI, productivity, writing, and future work styles, and one of the earliest teams to systematically integrate tools like Claude Code, Codex, and AI Agents into their daily work. Dan has been writing and researching how AI transforms knowledge work for a long time, offering many forward-looking insights based on his strong hands-on experience with AI tools, organizational collaboration, and changes in job roles.

⏱️ Timestamps

00:00 Opening & Episode Introduction

Why Dan can predict the future of AI

00:00 Chinese podcast opening: Cross-border visit plan and AI voice cloning explanation

00:37 Introduction of this cloned episode: Lenny’s Podcast and guest Dan Shipper

01:22 Preview of key points: AI will not bring a simple job apocalypse

03:01 Official introduction by Lenny: How Every lives in the future of work styles ahead of time

04:24 Review of the prescient prediction: How Dan foresaw the non-engineering potential of Claude Code a year ago

05:59 Every's work style: Making the entire team early adopters of AI

08:35 Three frameworks of predictions: work styles, job forms, and who will succeed

09:14 Retrospective after a year: Scoring these AI predictions

Two main threads of future work styles

09:57 AI Paradox: The stronger the model, the more work humans might have to do

10:55 First thread: Companies will first have a "super Agent"

14:09 Why Agents need to be continuously tended to like a garden

14:55 Slack will become an important entry point for work Agents

16:00 Second thread: Codex / Claude Code becoming the operating system for knowledge work

20:05 Major shift: It's not AI entering SaaS, but SaaS running inside Agents

21:30 Competition among Cursor, Claude, OpenAI, and Agent harnesses

23:11 How SaaS companies should redesign products for "human + Agent" collaboration

26:12 CLI era quickly passed: GUI + Agent will become mainstream again

27:49 Two Agents are better than one: Collaboration between personal Agents and application Agents

30:22 Counter-consensus view: SaaS won't die; Agents will increase SaaS user numbers

Why AI Automation Makes Humans More Important

32:09 Every's team doubled: Why an AI-native company is still hiring

33:00 "Automation is a lie": Each Agent needs a person

33:45 Senior Engineer Benchmark: The gap between AI programming and senior engineers

37:27 The real comparison is not AI vs human, but people using AI vs people using AI

38:50 First set of action recommendations: Use Codex more, make products usable by Agents, try company-level Agents

What Changes Are Happening in Job Forms

40:09 PR numbers surge: Non-technical people start submitting code and modifying products

42:11 Generalists become stronger: Market, operations, editors can do what used to be technical jobs

44:10 New job roles emerging: Forward-deployed engineers and Agent operators

45:14 New pain points for data scientists: Reviewing rough analyses done by others using AI

47:05 Which roles change the least: CEOs, managers, and sales

49:31 It's not about watching over Agents, but building systems that allow organizational capabilities to overflow

50:36 AI-generated documents and emails will be increasingly accepted

53:34 New uses for content: Written for both humans and Agents to read

Who Will Succeed in the AI Future

54:21 Dan is very optimistic about PMs: Product judgment will be amplified by AI

56:20 "PM is back, SaaS is back"

57:40 Opportunities for full-stack designers: From design ideas to directly submitting PRs

58:24 Creativity is more valuable: Standing out in the flood of AI-generated content

58:24 Job apocalypse will not happen easily: Models make "yesterday's human capabilities" cheaper

01:01:13 Most important career advice: Ride the models, follow the models

01:01:56 How to follow the models: Keep a playful mindset, repeatedly test new capability boundaries

01:04:40 True picture of the future: Everything has changed, yet nothing has changed

01:07:06 Final practice checklist: Move your workflow into Codex or Claude Code

Q&A

01:08:37 Enter Q&A: Scoring predictions after a year

01:08:50 Recommended reading list: *The Writing Life*, Churchill's WWII history, *The Rigor of Angels*

01:10:53 Recent favorite TV shows and movies: Knicks basketball, *The Dark Wizard*, *100 Foot Wave*

01:11:30 Recently favorite product: Codex

01:11:42 Life motto: Do things worth writing down, write things worth reading

01:12:45 Underappreciated AI Tools: Still Codex

01:14:27 How to Find Dan and Every, and Final Advice for Listeners

🌟 Highlights

💡 AI Paradox: The More Automation, the More Important Humans Become

Dan Shipper presents the counterintuitive insight of this episode: as models become stronger, work does not automatically decrease. Instead, humans will take on more tasks such as judgment, management, review, coordination, and creation. Agents can handle more and more execution tasks, but they need humans to define problems, maintain systems, judge the correctness of results.

"I have extreme faith in AI, but I also have great confidence in people. There is a lot of automation, but each Agent needs a person."

🤖 Future Work Will Be Divided into Two Categories: Agent Experiences

Dan predicts that the use of AI in knowledge work over the next year will primarily fall into two categories. The first category is super Agents within companies, possibly residing in Slack, handling data requests, business processes, internal knowledge tasks, etc. The second category is AI work environments like Codex, Claude Code, and Cowork running on personal computers, which will gradually become new work operating systems.

"Most of the work you do will actually happen on your computer, in environments like Codex or Claude Code."

📈 SaaS Won't Die; Agents Will Make SaaS Stronger

In response to the "SaaS Apocalypse," Dan offers a very contrarian view: SaaS will not be eliminated by Agents; instead, it will see increased demand due to Agent usage. In the future, not only humans but Agents will also be frequent users of SaaS. What SaaS companies truly need to do is not blindly integrate AI into their products but make their products suitable for collaboration between humans and Agents.

"I think the notion of a SaaS apocalypse is silly. If it were now, I would buy SaaS stocks. What Agents do is increase the number of SaaS users, not eliminate SaaS."

🧑‍💻 Why an AI-Native Company Like Every Is Still Hiring

Every doubled its staff last year, contrary to what many imagine for AI companies. Dan explains that automation does not completely remove humans from the system; instead, it creates new demands for management and maintenance. Each Agent needs someone to care for it, observe it, debug it, and ensure it is truly useful to the organization.

"Whenever you automate something, you need a person above to ensure the automation works well."

🛠️ From "Doing" to "Reviewing, Integrating, and Judging"

AI allows more non-technical people to write code, submit PRs, perform analysis, and generate documentation, but this also creates a large amount of output that needs to be reviewed and integrated. In the future, the focus of many professional roles will shift from completing tasks manually to judging what should be retained, how it should enter the system, what should be deleted, and how to maintain overall coherence.

"In the past, building things was very difficult. Now it has become very easy. So the focus is no longer on whether we can build it, but whether it fits with what we have already built."

🚀 PMs and Full-Stack Designers Will Be Amplified by AI

Dan is very optimistic about product managers and full-stack designers who truly know how to use AI. With lower barriers to entry, a PM's user understanding, product judgment, and storytelling skills will become even more critical; designers can turn their ideas directly into runnable products without relying entirely on engineering handoffs. Creativity will be more scarce in an era of AI-generated content.

"PMs and designers will do well."

🌊 Most Important Personal Advice: Ride the Models

Dan's final advice to everyone is: follow the models. Do not avoid new models out of fear; instead, use them in real work and life, continuously testing their new capabilities. He believes the frontier of AI is not just in San Francisco but wherever real people are using AI to solve real problems.

"The only thing you need to do is ride the models, which means following the models."

🌐 Podcast Information Supplement

This podcast uses the original human voice for audio production and may sound a bit strange in some places.

Translated using AI, so there might be some awkward phrasing;

If you want to listen to other foreign podcasts in Chinese in the future, feel free to contact WeChat: iEvenight