Untitled: A title-less future of work
- Rajib Ghosh
- Apr 25
- 9 min read
Who are you when the AI can do what you do?
You know this org chart. You live in it.
You're in a product company. There's a CPO at the top. Below them, product managers. Next to them, a Head of Design with designers, a VP of Engineering with engineers, a Head of Research with researchers. Everyone sits in a cross-functional squad, but everyone reports up through their function. Two bosses. Two ladders. Two cultures pulling in slightly different directions.
It's awkward. But it works. Sort of.
The whole thing rests on a single, unexamined assumption: design knowledge lives inside designers' heads. Engineering knowledge lives inside engineers' heads. So you group them together, give each group a leader who speaks their language, and call them departments.
Expertise is valuable. So you put the experts in boxes.
I spent 25 years in one of those boxes. I loved it there. I wrote about it last August, when I argued functional roles were safe. I wrote about it again a week later, when I described designers evolving into "system stewards." Transformed, but still designers. Still in the box.
Both articles were honest. Both were incomplete. Because I hadn't yet reckoned with a simple question:
What if the expertise isn't rare anymore?
The experiment that changed everything
In 2025, researchers at Harvard Business School ran a field experiment at Procter & Gamble. 776 professionals. Real product challenges. Real stakes.
Finding one: a single person with AI matched the output of an entire team without AI.
Finding two (the one that wrecked my sleep): without AI, the R&D folks proposed R&D solutions and the marketing folks proposed marketing solutions. Classic silo brain. With AI? Both groups produced balanced, cross-functional ideas. The functional identity just... evaporated.
If these expertise silos can have a very different shape with AI, we may want to rethink the design of organisations. Fabrizio Dell'Acqua, Harvard Business School, 2025
Read that again. The AI didn't just make people faster. It made the category of "R&D person" vs "marketing person" meaningless. The knowledge left the building. Or more precisely, it became available to everyone simultaneously.
If the expertise isn't locked inside people's heads anymore... why are we still organising around their heads?
But here's the thing most companies are getting wrong. They're trimming layers within departments. Fewer design directors, but still a design org. Fewer engineering managers, but still an engineering function. They're rearranging deck chairs on a ship that's being rebuilt from the hull up.
Worse: the halfway approach is actively backfiring. HBR documented a bank that gave its risk team one AI and its marketing team another. The risk AI flagged certain customers as dangerous. The marketing AI targeted those exact same customers for growth. Same bank. Opposite conclusions. Nobody noticed.
When every department gets its own AI, you don't get smarter. You get fragmented smart. The California Management Review put it plainly: AI agents can replace coordination roles, but only if you share them. Lock them inside departments and they just amplify the dysfunction.
The silo isn't just unnecessary anymore. It's becoming the thing that prevents AI from working.
Three things. That's all that's left.
Strip away the functional containers. Forget job titles for a minute. Ask: what do humans in a product company actually contribute that agents don't?
I keep landing on three things.
Scope. How big a problem can you hold in your head? "I own this checkout flow" is different from "I own the commercial strategy for the platform." That's altitude. Not function.
People. Trust. Growth. Feedback. Spotting burnout before someone says it. No agent does this. No agent should try.
AI fluency. Knowing when the Design agent is producing beautiful garbage. Knowing when the Finance agent's model is overfit. Knowing how to combine six agent skills into something that actually makes sense. This is the new expertise. The domain is the agent itself.
If these expertise silos can have a very different shape with AI, we may want to rethink the design of organisations. Josh Bersin, HR Tech Europe, 2025
McKinsey already describes two to five humans supervising 50 to 100 agents. MoonShot AI already operates this way. 300 people. No departments. No titles. Five founders, each overseeing 40 to 50 people directly.
If scope, people, and AI fluency are what matter, maybe build the org around those. Not around which course you took fifteen years ago.
Here's what it could look like.
And it's not just leaner. It's faster.
Here's something that gets lost in the philosophical argument about functions and identity: this model is dramatically more efficient. Not in the blunt "fewer humans, lower costs" way that got Klarna in trouble. In the "decisions that used to take weeks now take hours" way.
Think about how a new feature gets built in a traditional product company. The PM writes a brief. It goes to the design lead, who assigns a designer, who has their own sprint cadence. Design hands off to engineering, who estimates, pushes back, re-estimates. Research runs a study that takes three weeks. Finance weighs in on unit economics a month later. Legal reviews after that. Every handoff is a queue. Every queue is a delay. Every delay is a meeting to discuss the delay.
Now picture the same feature in the scope-based model.
Tuesday afternoon. You're a scope executor who owns the onboarding experience. You brief your Design agent and Engineering agent simultaneously. While they generate prototypes and technical approaches, your Research agent runs a rapid analysis of existing user data. Your Finance agent models three pricing scenarios. Your Legal agent flags a GDPR consideration you hadn't thought of. By Wednesday morning you have a prototype, a technical plan, a data-backed hypothesis, a financial model, and a compliance checklist. What took five teams and six weeks now took one person and eighteen hours.
The speed doesn't come from people working harder. It comes from removing the queues between functions. In a traditional org, the designer can't start until the PM finishes the brief. The engineer can't estimate until the designer finishes the mockup. Each function operates on its own timeline, with its own priorities, and its own backlog of other work. The calendar, not the problem, dictates the pace.
When one person orchestrates multiple agents in parallel, those queues disappear. The agents don't have competing sprint commitments. They don't need to context-switch from another product leader's project. They don't take PTO. The bottleneck shifts from "waiting for the next function to be available" to "how quickly can the human evaluate the output and make a call."
There's a compounding effect too. In the traditional model, if the financial analysis reveals the feature isn't viable, you've already spent weeks on design and engineering work that gets shelved. In the scope-based model, you learn that on day one, before significant effort is invested. The cost of killing a bad idea drops from weeks of sunk work to hours of agent compute.
This isn't theoretical. McKinsey's research on agentic organisations found that the length of tasks AI can reliably complete has been doubling every seven months since 2019, and every four months since 2024. They project AI systems could complete four days of unsupervised work by 2027. Pair that with a human who knows which four days of work to point them at, and you have something that's not just cheaper than a functional org. It's faster by an order of magnitude.
"Yeah but you still need experts to catch the AI's mistakes"
This is the best pushback I've heard. And it almost stopped me from writing this piece.
The worry: if you dissolve functional expertise, who catches the agent when it's wrong? A scope leader who knows nothing about finance rubber-stamping financial models isn't oversight. It's the Cigna scenario: a doctor approving 60,000 AI claim denials at 1.2 seconds each.
Fair. But this imagines the old version of judgement in a new context.
In the old world, judgement meant carrying enough in your head to spot problems as they flew past. The finance expert catches the flaw because they've seen similar ones over 15 years.
In the new world, you can ask.
Monday morning. You're leading the mobile product. You ask your Finance agent to model unit economics for a new feature. Instead of one spreadsheet, it returns three scenarios. Flags customer acquisition cost as the shakiest assumption. Shows you that if CAC rises 20%, the feature bleeds money in 8 of 12 months. It has already drafted a scaled-back alternative. You didn't need a decade in finance. You needed the right questions and the nerve to pick a path.
That's a different kind of judgement. Not "expert catches error from memory" but "decision-maker evaluates alternatives." That second skill is general. It requires critical thinking and comfort with ambiguity. Not a career in any single function.
And agents can pre-compute your plan B. In a traditional org, pivoting after a failed bet is expensive because it runs through separate functional pipelines. With agents, alternative plans are generated alongside the primary one. The cost of being wrong drops when your contingencies are free.
But what about Klarna?
They replaced 700 agents with AI. By 2025, the CEO admitted they'd over-indexed on efficiency. 55% of companies that rushed to swap humans for AI regretted it. But this isn't an argument against rethinking functions. It's an argument against lazy replacement. The model keeps humans. It changes what they're organised around.
And the thinking atrophy risk?
Gartner warns 50% of orgs will require AI-free assessments as critical thinking atrophies. This is real. It's exactly why AI fluency and people leadership sit at the centre. Judgement has to be actively built. Someone has to grow the next generation of sense-makers.
Does this work everywhere? No. Pharma compliance, safety-critical engineering, regulated finance probably still need deep specialists. But for most product decisions? Simulations plus the right questions plus good taste might be enough.
This is where I argue myself out of job. Or do I?
Twenty-five years in design. I've run teams, built systems, shipped products. Design is how I see the world.
In my August article, I described the designer becoming a system steward. Setting principles. Monitoring outcomes. Tweaking platform logic. That's still the right near-term move.
But follow the thread. If an AI agent can hold the design system, generate adaptive experiences, run experiments, and monitor quality metrics (all things I described the future designer doing), then "system stewardship" isn't a human role anymore. It's an agent capability.
What stays human? The judgement to know what experience is worth creating. The taste to feel when output is polished but soulless. The empathy to understand what data misses about how people feel.
Those are real skills. But here's the uncomfortable bit: they're not design skills. Not in any functional sense. The person who spots a lifeless interface from the Design agent is the same person who spots tone-deaf copy from Marketing or dodgy projections from Finance. The skill is cross-domain taste and judgement.
So yes. I might be arguing myself out of a title. But not out of work. The work moves from being a function to being a dimension of judgement every scope leader needs. Someone who spent 25 years developing design thinking, systems reasoning, and human empathy is pretty well equipped for that role.
They just won't put "designer" on their LinkedIn anymore.
How you get there without blowing everything up
You don't flip a switch. The dangerous period is the middle: agents deployed but functions intact. Leaders threatened. Agents siloed. Everyone half-fluent in the new thing and half-competent in the old one.
Start small. Pick one product team. Dissolve the functional labels. Share the agent pool. Let the squad figure out who owns what by scope, not discipline. Keep functional centres of excellence alive, but only as training infrastructure. The design expert doesn't vanish. They become the person who teaches everyone how to spot when the Design agent is wrong.
Then watch what happens.
The end state, if any of this holds, is a company where "What department are you in?" feels as dated as "Which typing pool are you assigned to?"
I could be wrong. AI adoption might slow. Functional expertise might prove stickier than I think. Coordination costs might not fall the way the models predict.
But the direction feels right. And I'd rather think about it now, while we can still shape it, than be surprised later.
Sources
Dell'Acqua, F. et al. (2025). "The Cybernetic Teammate." HBS Working Paper 25-043.
Korn Ferry (2025). Workforce 2025: Power Shifts. Survey of 15,000 professionals.
Revelio Labs (2025). 2025 Workforce Insights Wrapped.
Hoque, F. et al. (2026). "The Looming AI Risk." IMD.
Kenny, G. (2025). "Don't Let AI Reinforce Organizational Silos." HBR.
McKinsey (2025). "The Agentic Organization."
California Management Review (2025). "The Silo Effect in the AI Age."
HR Executive (2025). Josh Bersin on breaking down silos.
Fast Company (2025). AI and the death (and rebirth) of middle management.
Futunn (2025). MoonShot AI's departmentless structure.
Klarna (2024-25). CEO statements via various sources.
Comments