Knowledge Work Is Shallower Than You Think — AI Just Changed That

When Steve Ruiz started using Claude and other AI tools for his own work, he noticed something counterintuitive: he kept finding himself with extra capacity.

He’d finish a task, and instead of moving on, he’d think, “What if I took this deeper? What if I automated this part? What if I solved the version of this problem that usually isn’t worth solving?”

And then he realized: he’d been living in a world of necessarily shallow work forever. Not because the work should be shallow, but because doing it deep required hiring consultants, spending time, burning resources.

Now, with Claude 4.5 and similar tools, the constraint is gone. And the world isn’t ready.

The Shallow Work Baseline

Here’s how most organizations work right now:

You need a business decision. You could do deep analysis — pull data from five sources, model three scenarios, get input from stakeholders, do sensitivity analysis. That’s deep. It takes a week and costs $20k in consultant time.

Or you could do shallow analysis — look at the most recent report, have a meeting, make a guess. That takes an afternoon and costs nothing but uncertainty.

Guess what everyone chooses?

Same pattern everywhere. A startup wants to improve their marketing. Deep work: understand audience segmentation, test messaging with real users, analyze competitor positioning, iterate. Shallow work: copy what works on Product Hunt.

Education: a teacher wants personalized lesson plans. Deep: analyze each student’s learning pattern, design custom scaffolding, check for comprehension in real time. Shallow: one-size-fits-all lesson from a textbook.

That’s not laziness. It’s rational. If it takes 100 hours to do something deep, and you have 5 hours available, you do it shallow.

“A lot of tedious knowledge work, a lot of like things where, wow, if I had like 15 really smart consultants, like I could do a lot with that,” Ruiz says.

He’s describing the truth: organizations everywhere want to do deeper work. They just can’t afford it.

The Ceiling Is About to Break

AI doesn’t make knowledge work disappear. It removes the resource constraint.

A marketer with Claude can now do multi-stage prompting to test messaging variants, analyze them, iterate. Not in a week — in an hour. The depth that used to require consultants is now available to one person with a tool.

An educator with AI can generate personalized lesson plans for 30 students in a day, something that used to take a semester of planning time.

A founder with Claude and cursor can architect a system, build it, debug it, and think about optimizations — all in a timeframe that used to require a whole engineering team.

The work isn’t disappearing. The constraint is.

“I think what I’m recognizing is that like, wow, there’s actually a lot of work to do. Like there’s like, there’s a lot of stuff that like we were never trying to solve.”

This isn’t about working harder. It’s about finally being able to work on the things that matter.

The “God in a Box” Problem

Ruiz describes the instinct this way: “It’s like someone’s handed me just like God in a box. And I’m like, all right, well, how do I make it faster? How do I run two of these things at once? How do I optimize my God box in order to make this thing more powerful?”

Once engineers got access to Claude 4.5, they didn’t say, “Great, I can work less.” They said, “Great, what else can we solve?”

That’s the signal. When you remove a constraint, people don’t rest. They go deeper.

And the implications are enormous because they’re not just about speed. They’re about possibility.

Right now, your business decisions are informed by the data you had time to analyze. Your kids are educated by the lesson plans teachers could create in summer. Your product’s UX is informed by feedback that actually got collected and reviewed, not feedback that got lost.

All of that is shallow because of time constraints.

But what if every business decision was informed by deep analysis? What if every student got personalized education? What if product teams could test every hypothesis?

That’s not a small change. That’s a restructuring of what’s possible.

Why Organizations Aren’t Ready

The dangerous part of Ruiz’s insight is that it’s not obvious. Companies are already using ChatGPT and Claude. They’re using AI for coding, writing, research. They think they’re optimizing.

They’re not. They’re still operating in the shallow work paradigm. They’re using AI to do shallow work faster, not to do deep work.

Real readiness would mean:

• Changing processes to expect deeper analysis (which takes different kinds of thinking, different tools, different org structure)
• Training people to do deep work (some people are trained for shallow work and struggle with depth)
• Building feedback loops that actually use the deep work (an analysis that nobody acts on is just theatre)
• Accepting that depth is slower, even with AI (analyzing five scenarios takes longer than trusting your gut, even with AI helping)

“Our business decisions could have been better informed, our kids could have been better educated. Are politicians, could have had better data, everything that could benefit from essentially just applied engineered knowledge work will be done and there will still be more to do.”

That’s Ruiz’s optimistic thesis: the work is out there. The tools exist. But it requires a shift in how we think about what’s possible.

The Counterintuitive Implication

Here’s where it gets interesting: Ruiz isn’t arguing that AI will create 10x growth or 100x productivity in the conventional sense.

He’s saying something different: the ceiling of knowledge work is higher than we think, and we’ve just installed a longer ladder.

Whether we actually climb it is up to us.

Some organizations will use AI to do shallow work faster (efficiency play — let’s automate email). Some will do deeper work (effectiveness play — let’s rethink our email strategy).

But here’s the trick: you can’t see the difference from the outside. Both companies might use AI. One is in the shallow paradigm. One is in the deep paradigm.

The deep paradigm company will look like it’s working harder (because deep work is cognitively harder). It will also be solving problems the shallow company never even knew existed.

“I think what we’re gonna find is that we’ve actually been having a really shallow engagement with the amount of knowledge work that could be done historically and that it’s probably much, much, much deeper than we’ve thought about.”

What Happens Next

Ruiz’s closing vision is worth sitting with: if organizations actually embrace deep knowledge work, what changes?

Better business decisions → better companies → less failure Better education → better-educated kids → more options downstream Better analysis of policy → better policy → better outcomes

None of that is AI magic. It’s just work that was always worth doing but too expensive to do.

The scary version: organizations that don’t go deep get outcompeted by those that do. The gap widens. Shallow stays shallow.

The optimistic version: the tools are cheap and accessible. Any organization can access deep work now. The only barrier is willingness.

Ruiz is betting on the optimistic version. He thinks knowledge workers will see what’s possible and reach for it.

Whether he’s right depends on whether we collectively understand what he’s saying: AI hasn’t made work obsolete. It’s made deep work possible.

The real work is deciding to do it.

FAQ

Is “shallow work” just inefficiency we should eliminate?

Not exactly. Shallow work is often good enough and necessary to keep things running. The insight is that when you can go deep, you discover better solutions. The question is: what’s worth going deep on?

Aren’t people already using AI for deep work?

Some are, but most organizations are still using it for speed (do shallow work faster). True deep work requires different processes, different mindsets, different feedback loops. That’s the shift that hasn’t happened yet.

Can AI actually help with genuinely deep problems, or does it only work for routine tasks?

The transcript includes examples of both. Multi-stage prompting for marketing strategy is deep work with AI. AI agents self-organizing to visualize complex diagrams is deep work. Claude helping with firmware problems is deep work. The difference is asking the right questions.

Won’t “going deep” on everything just create overwhelming complexity?

Yes, which is why judgment matters. Not every decision needs deep analysis (choosing lunch doesn’t require consultants). The skill is knowing where depth matters. AI doesn’t remove that judgment; it just makes depth affordable when it does matter.

Is this different from how technology usually boosts productivity?

It’s similar but with a twist. Most tech improvements let you do the same work faster (email vs. letters). This removes a constraint (analysis, time, resources) and reveals what was always possible to do but too expensive. That’s a different category of change.

What if people just use AI to work less instead of going deeper?

Possible, and fine for individuals. But organizationally, competitors who go deeper will outperform those who just reduce effort. The market will enforce the deep work paradigm over time.

How does this apply to creative work?

Creative work is often deeper by nature (it requires imagination, iteration, taste). AI makes it possible to do more iterations, test more ideas, and refine more thoroughly. That’s deepening, not replacing.

Is Steve saying we should all be workaholics with AI tools?

No. He’s saying the capacity for deep work is now available, and we should choose consciously about where to apply it. That might mean working less on shallow stuff so you can focus on deep stuff.

What if AI stalls? Does this whole thesis fall apart?

If AI capability plateaus, yes. But Ruiz seems confident that improvements will continue (Claude 4.5, Gemini Flash, etc.). The trend is toward cheaper, better, more capable models. That reinforces the deep work opportunity.

How does this apply to my work right now?

Audit a decision or project you’re working on. Is it shallow (fast, good enough) or deep (thoughtful, thorough)? Could you go deeper with AI help? What would that unlock? Start there.