Changing the way you think about AI

Two articles stole the headlines this week in the discussion about AI and the future of work.

One argues AI is about to eliminate half of all entry-level white-collar jobs. The other found that people using AI aren't working less, they're actually working harder, longer, and burning out.

I've been using AI daily at beehiiv for months, and both articles are right.

Here's what I'm seeing, and what it means if you're a PM or founder who hasn't started yet.

The first article to take the internet by storm was Matt Shumer’s viral post: “Something Big Is Happening”.

This generated 8m views and caused quite a stir:

The core argument: AI capability is compounding faster than anything we've seen. Models went from failing basic arithmetic in 2022 to writing production software in 2024. The latest releases demonstrate genuine judgment, not just execution.

The headline stat: the CEO of Anthropic predicts AI will eliminate 50% of entry-level white-collar jobs within one to five years. Shumer thinks he's being conservative. METR's (Model Evaluation & Threat Research) data backs the pace. AI models doubled their independent task duration roughly every 7 months, from 10-minute tasks to nearly 5-hour tasks in a single year.

His practical advice is simple: pay for frontier models, spend an hour a day using AI on real work, and treat adaptability as your primary career skill. Nothing done on a computer is safe in the medium term.

The second: a report from HBR titled “AI Doesn’t Reduce Work—It Intensifies It”

Harvard Business Review published an eight-month study tracking what actually happens when people adopt AI tools at work. The finding contradicts the common narrative: AI doesn't lighten workloads. It amplifies them.

Three patterns emerged. First, task expansion. People absorbed work beyond their roles. PMs started writing code, researchers handled engineering tasks, and staff attempted work they'd previously outsourced. Second, blurred boundaries. AI's accessibility made it easy to fire off prompts during lunch, meetings, and downtime. Recovery time disappeared. Third, increased multitasking. People ran parallel workflows, managed multiple agents, and revived shelved projects simultaneously.

The result is a self-reinforcing cycle. Faster task completion raises speed expectations, which increases AI reliance, which widens the scope of attempted work. The initial productivity surge masks what's actually happening: workload creep, cognitive fatigue, and burnout.

The researchers' advice is to build deliberate norms around AI use. Intentional pauses, sequenced work instead of constant responsiveness, and protected time for human conversation. Without structure, "AI makes it easier to do more, but harder to stop."

1. AI is ready for everyday complex usage. Stop waiting.

The latest models from Anthropic and OpenAI aren't incremental upgrades, they represent a step change in reasoning and quality of output. As someone who has used every major model since GPT-3, the difference is obvious. These models consistently get it right first time. That shift has led me to trust AI with increasingly complex work at beehiiv, where it consistently delivers a similar quality output to me, but in a fraction of the time.

2. It's the worst (and most expensive) it will ever be.

At the current rate of change, Opus-4.6 and-(the latest and best models) will be obsolete in 6-9 months.

The model economics will have also changed: Mid-tier models in 2026 are 10x cheaper per token than frontier models in 2024. Even if AI isn't nailing your use case today or if the economics are not feasible, everything will change within two tech cycles. Start today, improve it tomorrow.

3. Output goes up. Work doesn't go down.

The HBR article highlights this idea of work intensifying - this is 100% accurate. AI compresses timelines across the board for PMs and significantly intensifies task velocity. Here are some examples of this for me last week:

Time per task collapses. Tasks that were previously impossible become feasible. The freed bandwidth doesn't disappear, it just gets filled with more work - this opens the door for entirely new wor,

4. Your role is shifting from creator to reviewer.

Developers already spend more time reviewing AI code than writing it. The same thing is happening to PMs. As you can see from the table above, the job is increasingly about directing agents, evaluating output, and knowing whether the result is good enough. The skill is becoming less about doing the work, it's more about knowing what to ask for, absorbing the information and being able to constantly context switch faster than you ever have before.

5. Role boundaries are dissolving.

The HBR study found that PMs started writing code, researchers handled engineering tasks, and people absorbed work they'd previously outsourced. AI lets you operate across traditional role boundaries and should open your aperture for what you can work on/solve yourself. If you’re a PM this is incredibly liberating.

6. The gap between adopters and non-adopters is huge and compounding.

Similarly to the differences that have always existed between PMs who could/couldn’t write SQL, AI creates a huge gap in leverage and output for PMs. Both skills created an unfair advantage in the workplace, however, the delta from AI adoption will be 10x greater.

7. You are still the bottleneck.

Three simultaneous agent conversations is the practical limit for most people, myself included. You can only review one document at a time, so no matter how fast your AI is, you are going to be the constraint. As a result, three personal abilities now matter more than ever:

Because AI makes almost anything possible, there is a high possibility of engaging in irrelevant/distracting work. These three traits will determine whether you can capture that value in this new world where you are no longer constrained by other people.

8. Start by asking questions, not giving instructions.

When I first started using Claude Code at beehiiv, the instinct was to build immediately. I’ve learnt that the better approach is spending your first sessions asking "how does this work?" and "explain this to me." Build understanding before you build anything. The quality of your output is directly proportional to your understanding of the tool, the context and the planning you do together.

9. Context is everything.

The single biggest factor in AI output quality isn't the model. It's the context you give it. At beehiiv, I drop a CLAUDE.md file into every project folder, essentially a briefing doc that tells the AI what the project does, what to avoid, and how things are structured. I also make sure that the instructions file addresses how to interact with various connectors and data sources (MCPs) such as linear, slack, unblocked, PRDs etc, which make sure it isnt working in a vaccum.

The difference in output quality is night and day. Think of it like onboarding a new team member. The better the briefing, the better the work.

10. Update your priors constantly.

Whatever you tried six months ago, try it again. The models couldn't do certain things a year ago that they handle effortlessly now. Your impression from a bad experience in 2024 is no longer valid. The capability curve is moving so fast that your mental model of "what AI can do" needs refreshing every few months. This is why regular and consistent use is essential.

11. Prompts are the new PRDs.

If you're building AI features and you're not prototyping with prompt sets, you're doing it wrong. The ability to define a clear outcome in natural language, to describe what success looks like so precisely that an agent can execute on it, is becoming one of the most valuable skills a PM can have.

12. PMs need to know more about coding instead of less.

There's a tempting narrative that PMs don’t need to understand code/the backend because AI handles the complexity for them. This is a mistake. Understanding how your product works technically is more important than ever, because you need to guide your agent and review their plan. And to do this you need to understand how the pieces of the product fit together, where things can break, and what tradeoffs matter.

13. Async agents drive huge time savings

One of the most underrated dimensions of AI agents is asynchronous work. You describe a task, walk away, and come back to a finished result. At beehiiv I try to kick off all my agent analysis and research projects first thing in the morning before I dive into meetings. As soon as the meetings are over I have all the analysis ready to review. Once again, this allows you to accomplish more and expand your capacity.

14. Your taste and your plan is your edge.

AI can generate an infinite supply of ideas, prototypes, and drafts. What it can't do is tell you which ones are good, how it aligns with the strategic business trade-offs or how it relates to the changing market dynamics. The editing function, knowing what to keep, what to cut, and what to push further, becomes the highest-leverage skill.

15. One hour a day. That's the minimum.

This is the #1 thing to take away from this newsletter - the only way to really get the most out of AI is by using it. To start, commit to one hour of hands-on usage, on a real problem, from your actual job. That's how fluency develops. I started with Claude Code by asking "how does this flow work?", “how would I automate this” and “can you do X?” Within weeks I was creating my own skills, saving 10+ hours every week and even teaching others. The compound effect of daily practice is enormous.

Thanks for reading - new content coming this Thursday!
(Read to the bottom to choose the next topic )
Jake

Why does business news feel like it’s written for people who already get it?

Morning Brew changes that.

It’s a free newsletter that breaks down what’s going on in business, finance, and tech — clearly, quickly, and with enough personality to keep things interesting. The result? You don’t just skim headlines. You actually understand what’s going on.

Try it yourself and join over 4 million professionals reading daily.

Check it out