An OTT Moment for Everyone: How Cowork Is Reordering the Applications of a Lifetime

Something fundamental has changed in the way I work.

When I need to start something, I no longer instinctively open Word, Excel or PowerPoint. I begin in cowork.

I may still enter the underlying application later — to inspect the result, verify a calculation, or test a possibility. But increasingly, I do not make the changes myself. I ask for them to be made.

This may look like a small change in habits. I believe it signals something larger.

For most of my career, knowledge work has been organised around applications. We learned to translate our objectives into documents, spreadsheets, presentations, emails and meetings. Collaboration platforms helped us exchange those artefacts, but human beings remained responsible for connecting them and carrying context from one to the next. Cowork reverses that relationship. It lets us begin with what needs to be accomplished, while the applications required are progressively activated underneath. They do not disappear. They stop being the centre around which the work revolves.

What I mean by cowork

I use cowork applications as shorthand for an emerging class of agentic work environments, not one product. Current examples include Anthropic's Claude Cowork, OpenAI's ChatGPT Work, Microsoft 365 Copilot Cowork and Google's Gemini Enterprise with Workspace Intelligence. Their architectures differ, but they share one trait: the knowledge worker expresses an objective, the system assembles context and works across application boundaries, and the human directs, inspects and approves the result.

What distinguishes these from earlier AI assistants isn't the conversational interface. It's the shift from answering to acting — from helping with one task to pursuing an objective across several tasks, sources and applications. The unit of interaction is no longer the prompt or the document. It is increasingly the outcome.

A professional life spent inside applications

The productivity applications that shaped my career were revolutionary — word processors, spreadsheets, presentation software, email, and later cloud platforms like Teams that made distributed collaboration possible at scale.

But they also trained us to organise work by the logic of the application. Explain something > open Word. Analyse something > open Excel. Persuade someone > open PowerPoint. Coordinate > send an email or book a meeting.

The interface was never neutral. It nudged the work into a shape before we had decided what the work needed to achieve. Over time, the line between doing the work and producing a representation of it blurred: the presentation was meant to support a decision, but preparing it could become the work itself; the spreadsheet was meant to illuminate the economics, but perfecting it could consume the attention that should have gone to the judgment behind it.

The declining return on PowerPoint time

I have probably spent more professional hours in PowerPoint than in any other application — first building decks myself, later reviewing decks built by others. Both consumed extraordinary amounts of time: another pass at the storyline, another simplified chart, another slide rebalanced for density.

Some of this was valuable. A strong deck can expose weak thinking and turn a complex analysis into something an organisation can act on. But I became increasingly conscious of the gap between the hours invested and the impact produced. I could measure the hours. I could see the slides improve. I could rarely establish that the resulting decision, or business outcome, improved by anything close to the same margin.

A pattern recurs across most board and investor decks I've reviewed: two working days spent on layout, colour and chart type for a recommendation a sharper page of prose would have carried in twenty minutes of reading. The polish arrived; the argument didn't get better. That gap — hours invested against decision quality moved — is the return I'm describing.

Opening PowerPoint also has a subtler cost: its visual environment competes for attention immediately. Layout, colour, spacing and structure all demand decisions before the underlying message is even settled — and a professional-looking slide can create a false sense of progress, looking finished while the thinking behind it is not. Cowork lets me delay that distraction.

Asking as thinking, applications as consequence

In cowork, I can start with the message, the objective and the audience — and discover, often, that my own thinking isn't precise enough yet. Reformulating the request, adding a constraint, challenging an assumption: this is itself a form of thinking. When a result isn't right, I have to work out whether the problem is in the execution or in how I framed what I wanted. Sharpening the request usually sharpens the thought.

This changes the organising logic of the work. In the traditional model, the artefact — the deck, the spreadsheet, the memo — is the immediate object: we decide we need one, then open the application and build it. In the cowork model, the artefact becomes a consequence of pursuing an objective. Preparing for a board meeting may require an analysis, a briefing note, a scenario comparison, a calendar action and a message to participants; instead of initiating and coordinating each element separately, the meeting itself — and the decision it must produce — becomes the organising unit, and the required artefacts assemble around it.

That matters because an objective rarely respects application boundaries. A customer problem doesn't divide neatly into a Word component, an Excel component and a Teams component — those boundaries belong to the software, not the problem. Cowork's biggest productivity contribution may not be that it performs each task faster, but that it removes the human effort of translating an objective into a sequence of application-specific tasks — a translation burden that has long been an invisible tax on knowledge work.

My own role shifts accordingly: from operating the application to directing, inspecting and challenging the work. I still need to understand a spreadsheet well enough to recognise an implausible assumption, or a deck well enough to detect when visual polish is disguising a weak argument. Delegating production increases, rather than reduces, the importance of informed inspection — a tension I return to later.

The hidden integration tax

The limits of the application model are even clearer in teams. Teams and Slack improved collaboration enormously, but their underlying model is still human-to-human coordination: people schedule the meetings, follow the threads, find the latest file version, reconcile contributions and update the deck when an assumption changes. Even when the information is technically available, a human remains responsible for integrating it.

This is part of why better connectivity hasn't made professional life feel more productive. As the cost of communication fell, its volume expanded — more messages, meetings, channels, files, notifications — without a matching gain in our ability to turn them into coherent action. Much of what we call collaboration is really coordination overhead: summarising what others said, reconciling versions, briefing latecomers, translating one team's specialist output into another team's language. Productivity applications helped us create the pieces; collaboration platforms helped us exchange them; humans still had to assemble the whole. Cowork begins to absorb part of that tax.

Teams connects people; cowork connects contributions

Traditional platforms create spaces where people communicate. Cowork can become an active participant in what those people are trying to achieve — relating a customer email to a change in requirements, a comment to a challenge against an assumption, a delayed action to a project risk.

This changes collaboration's organising unit, from containers (a meeting, a channel, a folder) to the objective itself — assess an investment, respond to a customer problem, reach a decision. The workstream becomes a persistent, active body of context: what's been established, where contributors disagree, which assumptions remain unsupported. Contributions no longer need a long chain of handovers, and colleagues joining later need less of a manual reconstruction of the project's history. Some of the administrative work project leads currently do — consolidating inputs, maintaining the narrative, chasing actions — moves to cowork, freeing the team to spend more time on framing the problem and making the decision.

Both OTT and overbuild

Two patterns from telecommunications map onto this well.

The first is over-the-top disruption. WhatsApp didn't replace telecom networks — it used them while taking control of the customer's service experience. Users thought about their conversations, not the infrastructure carrying them; the network kept the traffic while losing much of the customer relationship. Cowork produces a comparable effect above productivity applications: Word still renders the document, Excel still runs the calculation, but cowork increasingly receives the user's intent and presents the integrated result. The legacy applications may keep the workload while losing part of the user relationship.

But cowork is also beginning to overbuild the fabric underneath, the way fibre doesn't just improve copper but creates new infrastructure with different economics — even while the old infrastructure keeps running for years. Cowork is starting to build a new fabric across applications: retrieving context, connecting contributions, detecting contradictions, requesting approval, preserving continuity between stages. It captures the eyeballs from above while overbuilding the digital fabric from within.

A second parallel: banking

Banking offers an equally close parallel, and it shows the same two-sided pattern from a different angle.

At the customer-facing end, wallets, neobanks and fintech apps inserted themselves between the customer and the traditional bank, competing for the interface, the daily engagement and the service experience. At the infrastructure end, open banking APIs, banking-as-a-service, embedded finance and modular payment services began decomposing the vertically integrated bank from within. The licensed institution and its ledger remained essential — someone still has to hold the account and settle the transaction — but they could become nearly invisible. A customer might use a financial service through a retailer or mobility platform without ever consciously touching the bank behind it.

The mapping onto cowork is close: the productivity application is the licensed bank and its ledger — necessary, regulated, and increasingly out of view; cowork is the wallet or fintech interface that owns the relationship; connectors and agent protocols play the role of open banking APIs; agentically assembled workflows resemble embedded finance, service quietly stitched into someone else's context.

The lesson from banking is the same one telecom teaches, stated more starkly: an incumbent can keep most of the regulated, load-bearing value — the ledger, the compliance, the underlying execution — and still lose a disproportionate share of strategic influence to whoever controls the interface and the orchestration around it. Holding the infrastructure is not the same as holding the relationship.

When coordination is also a source of power

Reducing the integration tax is not a politically neutral development. Many organisational roles carry a significant coordination load — collecting updates, consolidating reports, preparing the status summary, controlling the meeting agenda — and coordination is also a source of influence. The person who collects the information shapes how it's interpreted; the person through whom workstreams connect can become indispensable because no one else holds the full picture.

If cowork can maintain context and make the state of the work continuously visible, it changes who controls information, not just how much time is saved — and this will bear most directly on the parts of middle management built around collecting, synthesising and transmitting status rather than around judgment, coaching and accountability, which should matter more, not less, in an agentic organisation. Expect some resistance to surface as concerns about quality or governance even when the underlying issue is a loss of informational control — though not every such concern will be a disguise; many will be legitimate, and organisations should not expect this transition to be frictionless.

Why there will be no easy return

Despite that resistance, I find it hard to imagine a broad reversion to application-centred work once people have experienced work organised around an objective — not just because cowork is faster, but because the old model starts to feel wrongly designed. Once you're used to stating an objective, reconstructing a workflow application by application feels artificial.

The effect also compounds: cowork becomes acquainted with a team's terminology, stakeholders, prior decisions and approval habits, so its value sits as much in accumulated context as in the underlying model. Eventually, organisations will start removing the routines built to compensate for fragmented information — status meetings that exist mainly to transfer information, manually prepared progress reports, repeated briefings, version-reconciliation exercises. At that point cowork will have helped redesign the operating model, not just added a tool to it. The old applications will remain; the team will no longer revolve around them.

What becomes of the application interface?

If cowork becomes the centre of knowledge work, will Microsoft, Google and Apple keep investing at the same intensity in Word, Excel, PowerPoint, Docs and Sheets as self-contained destinations?

I doubt these interfaces disappear, but I expect their purpose to shift — from places where humans create things to places where humans inspect, interrogate and occasionally manipulate what's been produced elsewhere. I may not build the spreadsheet, but I still need to trace its formulas and test its sensitivities; I may not construct the deck slide by slide, but I need to verify its evidence and spot where polish is covering an unresolved argument. The design challenge moves from enabling production to enabling informed inspection and control — and applications will increasingly need to serve a second user, the agent, whose actions must be machine-legible and leave enough evidence for a human to verify what happened. Software companies will need to design two experiences at once: execution for the agent, inspection for the human.

From clicks to intent

Cowork also gives platform providers a different kind of data. Conventional telemetry shows behaviour — what was clicked, which feature was used. Cowork data can show purpose — what someone was trying to do, why the first result wasn't good enough, which assumption got challenged, which action got refused. Traditional telemetry observes manipulation; cowork can observe intent, correction and judgment.

That data could help decide not which toolbar to redesign, but which interactions no longer need to happen at all, and where human approval genuinely still matters. It's an extraordinarily valuable dataset — and a sensitive one, since it captures how people formulate problems and exercise judgment, not just how they click. That raises real questions about privacy and how much professional context a small number of platform providers should hold.

Will applications lose the human touch?

Traditional interfaces support more than functionality — moving elements on a slide can surface a relationship you hadn't noticed; writing and rewriting a paragraph by hand can expose gaps in your own thinking. Direct manipulation builds familiarity with the material. If cowork mediates every interaction, that familiarity can erode, and the software itself loses some of its authored, human character as an agentic layer generates a different path for every user rather than one coherent designed experience. The paradox: cowork can make software more responsive to human intention while making the application itself less directly human. The answer isn't a binary choice between manual work and invisible automation — it's a deliberate movement between delegation and direct engagement. Cowork should do what doesn't deserve my attention; the application should let me intervene where my attention adds value. I can delegate manipulation. I cannot delegate accountability.

The inspection paradox

This movement from production to inspection sounds efficient, but it creates a real difficulty: inspecting complex work someone else produced can be harder than building it yourself. The person who builds a model acquires its assumptions and compromises as they go; the reviewer receives the finished structure and has to reconstruct that reasoning backwards — the same problem software engineers know from code review, where reading unfamiliar code is often harder than writing it.

There's a longer-term version of this risk, too. Informed inspection depends on domain expertise, and expertise is partly maintained through practice. If I stop building the models and structuring the decks myself, will I still be able to inspect them with real depth in five years? I've written elsewhere about the Verification Tax — the per-task cost of checking what a machine produces. What I'm describing here is closer to verification debt: not the cost of checking any single output, but the slow depreciation of the judgment you'd need to check it well, accumulating exactly as the organisation grows more dependent on the system it's less equipped to challenge.

The answer isn't to insist humans keep manufacturing everything by hand — that gives up much of cowork's value. It's to deliberately preserve enough direct practice and independent reconstruction to keep judgment sound. The future knowledge worker may not need to manufacture every artefact, but must remain able to reconstruct its critical elements when the stakes call for it.

A new control model for teamwork

This paradox matters more in teams. Cowork can produce a coherent result before everyone involved has an equivalent understanding of how it was reached; fluency can make an argument look more robust than its evidence, and a misunderstood source or unstated assumption can enter a polished document and anchor the whole team's thinking. As the cost of producing an answer falls, the value of knowing whether the answer can be trusted rises — and responsibility can blur when several people and an agent have all contributed. "Cowork produced it" cannot be an acceptable answer to "who is responsible for this."

That calls for a clearer distinction between an output that has been produced, one whose material claims have been verified, and one a named person has approved and is prepared to defend. Important claims should stay traceable to evidence, explicit assumptions or named judgment; consequential outputs need identifiable owners; and at critical moments, teams should reconstruct part of the reasoning independently rather than asking the same system that generated the answer to confirm itself. Provenance, ownership and decision rights become more important, not less, and the time recovered from production needs to be consciously reinvested in challenge and verification.

The danger of industrialised output

One last risk: cowork can industrialise the production of artefacts nobody needs. If organisations keep measuring productivity by volume — more decks, more reports, more messages — AI will simply help them generate more polished noise. The recipient gets a cowork application that summarises the unnecessary deck; the sender gets one that expands the summary back into a new deck; two agentic systems compress and expand information while the humans drift further from the original purpose. That's automation without progress.

Cowork's value shouldn't be measured in hours saved or artefacts produced, but in whether it improves the quality and speed of decisions, shortens the distance between insight and action, and frees human attention for the judgment calls that actually matter. The real question isn't how much time cowork releases — it's what we do with the attention it returns.

Returning attention to impact

Productivity software helped us produce more. Collaboration software helped us exchange more. Cowork organises both individual and collective effort around what needs to be accomplished — that's the shift.

It doesn't eliminate documents, spreadsheets, decks, messages or meetings, or the expertise behind them. It changes their position: applications become instruments rather than destinations, artefacts become means rather than ends, and human attention can move from the mechanics of production toward purpose, judgment and verification.

None of this is automatic. Cowork can return attention to impact, or multiply output; strengthen collective intelligence, or produce collective confidence without understanding; free managers from coordination overhead, or provoke resistance from those whose authority depended on controlling information. The technology changes what's possible. What the recovered attention is used for is still our decision.

Just as an OTT user thinks about the conversation, not the network carrying it, the cowork user can increasingly think about the outcome, not the applications producing it.

For me, that's the real promise — not that I'll make more presentations, or even make them faster, but that I can spend less of my professional life perfecting representations of work, and more of it determining whether the work will make a difference.