AI Leverage and Judgment: The Future of Work

AI leverage and judgment infographic from Bora's UCLA Anderson talk

A room full of undergrads will tell you the truth faster than a room full of executives.

Execs have too much to protect. Titles. Budgets. Org charts. The comforting fiction that the machine they built only needs a few AI upgrades.

Students have less to defend. They’re walking into the blast radius. That makes them a better signal.

This week, my friend Ramit Varma, Founder and CEO of Breakout Learning, invited me to speak to the Business and Entrepreneurship class he teaches at UCLA Anderson. The topic was AI, but the real conversation was work. What happens to it. Who does it. Who gets paid for it. What skills still matter. What kind of company you should start when the customer no longer behaves the way the internet trained us to expect.

I brought my son Vuk with me. He just finished his second year in York University’s BDes program in Toronto.

I was speaking to a room full of students entering the same future my own son is walking into.

Bora Nikolic speaking about AI leverage and judgment at UCLA Anderson

That changes the weight of the conversation.

You stop trying to sound smart.

You try to be useful.

And useful, in this moment, means being honest about the thing AI makes harder to fake: judgment.

Because once leverage becomes cheap, judgment becomes the test.

The AI conversation is still pointed at the wrong target

Most AI conversations still orbit the same tired questions.

Will AI replace jobs?

Will students cheat?

Will software engineers survive?

Will designers still matter?

Will consultants have anything left to bill for?

Fine. Those questions matter. They also keep us staring at the worker while the customer changes shape.

The bigger shift is this:

AI changes who does the work.

A customer still wants the outcome. A ticket. A trip. A refund. A product. A recommendation. A decision. A support issue solved before it becomes a full emotional event.

The agent does more of the work in the middle.

Search. Compare. Read. Click. Wait. Fill forms. Chase support. Interpret policies. Calculate total cost. Find the better option. Ask for approval. Complete the transaction.

I wrote a deeper version of this argument in AI Agents Are Changing the Future of Digital Business, where the core point was that the next customer of your digital product may never move through the journey the way your team designed it.

A lot of companies have spent years confusing customer effort with engagement.

Clicking through six screens to cancel a subscription is customer labor. Waiting on hold is customer labor. Reading a policy written by legal and disguised as UX copy is customer labor.

Customer labor becoming agent work in AI-driven business

Agents are going to make that harder to hide.

The customer journey is becoming agent-mediated

I asked the students to imagine their AI agent planning their Friday night.

It finds the event. Picks the ticket. Compares prices. Books the ride. Makes dinner plans. Splits costs with friends. Handles changes if something goes sideways.

Then I asked who would use it.

Hands went up.

Of course they did. Normal people don't wake up excited to compare ticket prices, parking options, refund policies, group texts, dinner times, and ride-share prices. They want the night out. The work around the night out is mostly administrative sludge.

That hand raise contains a massive business problem.

If customers begin delegating that work to agents, every company has to answer a new question:

Why would an agent trust your business enough to act?

That question is far more useful than most redesign briefs.

A human might forgive friction because the brand is familiar, the habit is strong, or the alternative feels annoying. An agent will compare. It will inspect the details. It will expose unclear pricing, weak policies, broken support flows, bad data, and terms that only work because humans are too tired to read them.

A brand promise has to become system behavior.

Trust moves from marketing language into operational infrastructure.

AI agents need operational trust through data pricing policy and support

Accurate data. Clear pricing. Clear policies. Permissions. Security. Reliable support. Auditability. Consistent business rules.

The surface still matters. Digital platforms and touch points still matter. The experience still matters.

The underlying business systems that compose the engine matter more.

The surface is the wrapper. The engine is the business.

Digital platforms and customer touch points are the visible layer.

The real business sits underneath.

Business engine underneath digital platforms and touch points

Customer records. Pricing. Payments. Inventory. Policy. Loyalty. Permissions. Support logic. Rights. Content rules. Approval thresholds. Data quality. Operational discipline.

Those systems need to expose what matters and interact cleanly with both humans and agents. As agents proliferate, that interaction will likely shift toward agent-first. The customer may still be human, but the first operational contact with the business may increasingly be software acting on the customer's behalf.

That is where a lot of digital strategy is about to get humbled.

A lot of digital transformation was surface renovation. New wrapper. Same plumbing. Better interface on top of bad policy. Better content on top of broken data. Better checkout on top of confusing pricing.

Agents will see right through that.

When the customer journey becomes mediated by software that can inspect options, compare terms, and act across systems, the underlying business has to become more legible.

The wrapper no longer gets to carry the whole story.

The product cycle is collapsing, and students already know it

The second big theme of the talk was product creation.

AI is compressing the distance between idea and output.

AI judgment turns product cycle compression into value

The old product-making cycle moved through strategy, research, design, product planning, engineering, QA, security, launch, learning, and iteration. In healthy organizations, those things already overlapped. In unhealthy organizations, they became a slow-motion handoff parade.

AI is pulling that cycle tighter.

A sentence can become a prototype. A prototype can become code. A customer insight can become a testable concept. A rough product idea can become something concrete enough to debate.

The earlier version of this thesis showed up in Vibe Everything, where I argued that strategy, design, and build would collapse into one loop. I expanded the operating version in I Predicted The Product Cycle Would Compress. I Was Too Conservative, because the compression is happening faster than I expected.

That changes the economics of starting.

It also changes the standards for surviving.

I asked the room who had used AI to create a first draft of something that quarter. Hands went up again. Writing, research, slides, notes, studying, maybe some code. I didn't need a survey. The room told the truth.

Then I asked how many of those first drafts still needed real judgment.

Hands stayed up.

That is the whole AI productivity conversation in one gesture.

AI gives you output. Judgment turns output into value.

The first draft is losing its value as a protected activity. That applies to writing, design, code, research, strategy, marketing, analysis, and probably half the things sitting inside a normal white-collar workflow.

That is also why Claude Design matters. The first credible artifact is no longer protected territory, and that moves value toward orchestration, standards, and judgment.

The value moves to knowing what should survive.

What is true?

What is useful?

What is shallow?

What is risky?

What is missing?

What deserves to ship?

What should be killed immediately before it wastes everyone's time?

AI creates more options than people can use. The new bottleneck is selection.

When options get cheap, judgment gets expensive.

Starting gets easier. Standing out gets harder.

Students understand the startup angle immediately.

AI lowers the cost of starting. You can prototype faster. Write faster. Research faster. Design faster. Build faster. Launch faster. A small team can now create output that used to require a bigger team, more money, and more coordination.

That sounds like pure upside until you take the next step.

If starting gets easier for you, it gets easier for everyone else too.

The market gets louder. The number of credible-looking ideas explodes. More products look polished. More pitch decks look competent. More demos look real. More founders can fake early momentum long enough to get a meeting.

So the advantage moves.

It moves toward customer insight. Distribution. Trust. Timing. Taste. Operations. Cash discipline. Resilience. Judgment.

Speed helps. Speed alone won't build a company.

A weak idea moving faster is still a weak idea. Now it just wastes less time before reality punches it in the face.

Agents are the bill

The students also talked about gross margin in their pre-work, which made the economics section land better.

AI feels cheap when it's one person using one tool.

Agentic AI gets expensive when it starts doing real work.

A simple answer costs one thing. An agent completing a workflow costs something else.

It needs context. It uses tools. It retries. It checks data. It asks for approval. It monitors results. It logs actions. It needs security. It needs governance. It needs people who understand how to design, test, and control the system.

The cost doesn't disappear. It moves.

From headcount to compute. From manual work to orchestration. From software seats to usage. From production labor to review and governance. From slow coordination to system design.

The CFO is going to discover that AI productivity has a usage meter attached to it.

A company may remove labor from one line item and create a new tax on every workflow.

Agents are the bill.

The leadership question becomes painfully simple:

What outcome pays for the cost?
Does it reduce support cost?
Increase conversion?
Improve retention?
Speed up product learning?
Reduce rework?
Improve quality?
Open a new revenue stream?

I have written before that design and product work only matter when they connect to business outcomes. AI makes that even more direct. If the outcome is vague, the AI budget is fiction.

Every company is a factory

I have been using a factory analogy inside our own company because it makes AI transformation easier to plan.

Every company is really a network of internal factories.

Strategy turns ambiguity into direction. Product turns direction into priorities. Design turns priorities into experiences. Engineering turns experiences into systems. Operations keeps the machine moving. Commercial brings market signal back into the system. Support shows where reality broke the plan.

Each factory has inputs, outputs, standards, handoffs, rituals, and tribal knowledge.

AI judgment in AI transformation and factory redesign

Clients and customers often think they are buying the visible output. A digital platform. A product strategy. A redesign. A support model. A new fan experience.

What they are really buying is access to the factory system behind it.

Coordinated expertise. Process. Judgment. Context. Delivery discipline. The ability to move a problem through a network of specialized teams and turn it into something real.

This is why I keep coming back to the relationship between design and engineering. I wrote about it in Investing in Design and Engineering, but the AI version raises the stakes. The handoff can no longer be where quality goes to die. The factory has to work as a connected production system.

AI changes the machinery inside each factory.

That is the real transformation.

The obvious question is: how do we use AI inside the current workflow?

The better question is: what should this factory become when AI changes the cost, speed, and shape of production?

That lens makes transformation easier to operationalize because it forces better questions.

What should people still create directly?

What should be automated or semi-automated?

What still requires judgment, taste, review, and human accountability?

Where do handoffs kill speed?

Where does quality drop?

This is where AI transformation gets real.

Tool adoption is easy. Work redesign is the actual fight.

The future of work is exciting and brutal

This was the part that mattered most because the students in that room are walking into the labor market AI is reshaping.

There is real upside.

More leverage. Faster learning. Smaller teams. New companies. Less repetitive work. More people able to make things earlier in their careers. A student can prototype something that used to require a full team. A founder can test an idea before spending a year begging for resources.

That part is exciting.

The harder part is just as real.

Fewer entry-level paths. Displaced roles. Skill erosion. Identity shock. A lot of people learn by doing junior work. If AI absorbs too much of that work, the path to expertise gets more complicated.

Then there is the emotional layer.

People are going to grieve parts of the craft.

We keep talking about job loss like it is only an income problem. It is also an identity problem.

Work is how adults measure competence, status, belonging, usefulness, and sometimes self-worth. Strip away parts of the craft and you are changing someone's story about themselves.

People chose careers because they loved building software, designing products, writing sentences, analyzing markets, researching problems, solving messy puzzles, producing the thing with their own hands and mind.

When AI starts doing parts of that work, the reaction won't only be rational.

It will be personal.

The work changes before people are emotionally ready.

That is where a lot of leadership will fail. Leaders will talk about efficiency while people are processing identity loss.

Bad move.

The companies that handle this well will treat AI transformation as a human operating model shift, not a procurement exercise.

What do people get paid for?

This was the question I kept coming back to with the students.

If production gets cheaper, what do people get paid for?

My answer:

Accountability.

Judgment. Taste. Trust. Leadership. Relationships. Ownership. Systems thinking. Making hard calls. Knowing what matters. Knowing when a machine is wrong. Knowing when a team is lying to itself. Knowing when the obvious answer is expensive garbage in a nicer wrapper.

Technical skill still matters. Depth still matters. You need to be good at something real.

The strongest people will have depth and range.

I built my own career around that model.

The phrase people use is T-shaped. I think of it more as a generalist-specialist. It sounds like a contradiction. It is really a productive paradox.

AI judgment and T-shaped skills for the future of work

Go deep enough in one area to have real craft. Get broad enough around it to understand how the work connects to the business.

That combination changed my career. Engineering gave me the deep system foundation. Design, product, entrepreneurship, leadership, and sports technology gave me the range to connect the work to people, markets, teams, and outcomes.

I wrote about this in Why Being Great at One Thing Isn’t Enough Anymore, and the point has only gotten sharper. Depth gives you authority, but range gives you leverage.

I use the same model with my teams.

From a leadership perspective, it is one of the ways I think about learning and development. Where is someone deep? Where are they too narrow? What fluency do they need around their core skill to become more valuable over time? How do we upskill the team without turning everyone into the same generic AI operator?

That matters more now.

T-shaped skills for AI leverage and judgment

Breadth without depth becomes shallow.

Depth without fluency becomes fragile.

Don't use AI to avoid the work

This is what I wanted the students to leave with.

Use AI to make yourself more dangerous.

That means using it to build things, test ideas, learn faster, ask better questions, understand business, pressure test assumptions, and explore the parts of a problem you used to avoid because you lacked the skill, time, or confidence.

It does not mean outsourcing the hard parts of thinking.

If AI becomes your way around thinking, you are training yourself to be replaced.

Use it to get deeper into the problem. Use it to see more angles, pressure test more assumptions, and build faster feedback loops. The tool is useful. The work still belongs to you.

That distinction will matter.

A lot of people will use AI to produce more noise. Some people will use it to make better decisions.

Those are very different futures.

The room gave me hope

The UCLA students were engaged. They asked real questions. They participated. They pushed on the parts that mattered: work, trust, startups, skill, and what their own future might look like.

That part stuck with me.

It is easy to sit inside executive conversations about AI and drown in strategy language. Operating models. Margin pressure. Agentic workflows. Product-cycle compression. Governance. Infrastructure. Platform shifts.

All of that matters.

Then you stand in front of students and the question gets much simpler.

What should I learn?

What should I build?

What should I trust?

What happens to work?

Where do I fit?

Those questions are cleaner. They are also harder to dodge.

I looked at that room, and I looked at Vuk sitting there as a second-year university student, and the whole thing felt less theoretical.

This is the generation that has to turn AI from novelty into capability.

They will inherit the leverage.

They will also inherit the mess.

AI leverage and judgment will separate winners from noise

AI gives leverage.

That part is already here.

It gives leverage to the serious student and the lazy student. The disciplined founder and the delusional founder. The strong company and the broken one. The thoughtful leader and the executive who just wants to tell the board they have an AI strategy.

Leverage is no longer the advantage.

Leverage is the new baseline.

The advantage is judgment.

Knowing what to build. Knowing what to ignore. Knowing when the machine is wrong. Knowing when speed is hiding stupidity. Knowing when a polished answer is still empty. Knowing when the customer problem is real. Knowing when the economics work. Knowing when the system is safe enough to trust.

That is the uncomfortable part of the AI era.

AI will make mediocre thinking look more impressive.

It will make weak ideas look more fundable.

It will make bad strategy look more articulate.

It will make average performers look busy.

For a while, that will fool people.

Then reality will collect.

The future belongs to people who turn leverage into judgment.

Everyone else will confuse output with value.

AI Leverage and Judgment: The Future Belongs to People Who Decide