It's a deliberately provocative line, so let's get the obvious objections out of the way before the argument starts.
No, Claude is not literally a web browser. No, this is not a prediction that one company wins everything. No, "over 50% of what we do on a computer" is not a hill I'd plant a flag on in a board paper next week. It's a thesis about the direction of travel, not a forecast with a date on it.
The claim, stated plainly: the AI assistant — Claude, in our case — is on track to become the primary surface through which people use computers for knowledge work, in the same way the web browser became the primary surface for using the internet. Today that feels like a stretch. So did "everything will run in a browser" in 1998.
The browser took longer than people remember
It's easy to forget how long the desktop era lasted after the browser arrived.
Mosaic shipped in 1993. Netscape Navigator hit the market in 1994. By 1998, most office workers had used a browser — but the things they spent their day in were still desktop applications. Word. Excel. Outlook. Lotus Notes. Photoshop. SAP GUI. The browser was a place you went to look things up, not a place you went to work.
The transition didn't happen in one step. It happened in waves, each one looking small at the time:
- Read-only web (mid-90s): a publishing surface. The browser was a magazine rack.
- Webmail (Hotmail '96, Gmail '04): the first daily-use productivity tool to credibly move into the browser. People still kept Outlook open alongside it for years.
- SaaS proper (Salesforce '99, Google Docs '06, Workday '06): line-of-business software you used to install, now delivered through a URL.
- The browser as OS (Chrome '08, Chromebook '11, Figma '16): tools that were better in the browser than they'd ever been on the desktop. Figma killing Sketch is the moment the trend became undeniable, and that was twenty-three years after Mosaic.
At every stage there were sensible people pointing out that the browser was slow, the network was unreliable, the security model was wrong, and serious work would always live on the desktop. They weren't wrong, exactly. They were describing the present accurately and assuming it would stay that way.
What changed wasn't the browser. It was what got built behind it.
The browser of 1998 couldn't run Figma. Not even close. What made the browser the dominant work surface wasn't that browsers got faster (they did) or that networks got better (they did). It was that an entire generation of software got rebuilt with the browser as the assumed front end. The control plane moved. Once the control plane was the URL bar, everything else followed: auth, collaboration, distribution, deployment, even the way teams were organised.
This is the part of the analogy that matters for AI. The question isn't "is Claude good enough to replace Excel today?" The question is: if you were building a new piece of line-of-business software in 2027, would you assume the user's primary interaction with it is a chat thread with an assistant that can drive your app on their behalf? A growing number of teams already are. Not all of them are right. Enough of them are right that the gravitational pull is real.
Take a renewals analyst at a brokerage. Today they open the broking system, pull the schedule, cross-check the binder against the client file, draft the renewal pack in Word, send it from Outlook. Tomorrow they ask Claude to pull together the Q3 renewals for a named client, and the assistant operates across the broking system, the document store and the email outbox on their behalf — the analyst reads, edits, signs off, sends. The apps are still there. The centre of gravity has moved.
The browser-wars analogy, with the obvious caveats
If we lean into the analogy — and analogies are tools for thinking, not predictions — here's a rough mapping. Treat it as one person's read on where things sit today. The market is moving fast enough that any of these names could be swapped in or out by next quarter, and the mapping itself is a bit forced — I'm reaching for the closest historical shape, not claiming the fit is clean:
- Microsoft Copilot is Internet Explorer. Bundled with the OS, distributed by default, owned by the incumbent. Wins enormous share by virtue of being already-there. Not where the people who care about the craft want to spend their day.
- Claude is Chrome. The one that quietly wins the people who actually do the work, because the product is better at the thing it claims to do. (We obviously have a horse in this race. The rest of the post is the argument; the disclosure is the disclosure.)
- ChatGPT is Firefox. First mover, beloved, the one that proved this category existed at all. The reason any of the rest of this conversation is happening.
The analogy breaks in places — that's fine, all analogies do. Firefox never had the brand recognition ChatGPT does. Internet Explorer never had a generative model under the hood. Chrome didn't carry the safety and alignment baggage Claude does. The point isn't that history is repeating note for note. The point is that the shape of the market — an early experimental winner, a default-bundled incumbent, and a credible challenger that wins on craft — is a shape we have seen before, and we know roughly how it plays out.
It plays out over a decade or more, with several reversals along the way, and the eventual winner is rarely the company that looked dominant in year three.
What "the control plane" actually means
When we say AI becomes the control plane, here's what we mean concretely, in the order it's likely to land for a knowledge-work team:
- The assistant is where you start the task. Not the application. You don't open Excel to build a model; you ask the assistant to build a model and Excel is one of the tools it reaches for. The app is downstream of the intent.
- The assistant has memory and context the apps don't. It knows what you were working on yesterday, who you're writing this email for, which client this matters to. The apps are stateless tools it composes.
- The assistant operates the apps on your behalf. It connects to systems that have proper APIs through MCP servers — the emerging standard for plugging an AI assistant into enterprise software — and drives the systems that don't through computer-use, where the assistant operates an app's UI directly, the way a person would. The human dropping down into the actual UI becomes the exception, not the rule.
Each of these is happening at different speeds in different teams. Some have been doing all three for a year; some haven't started. None of it is universal.
Some categories will resist for a long time (anything where direct manipulation is the point — design tools, CAD, video editing). Some categories will flip fast (anything where the UI was always a clumsy proxy for "tell the computer what you want it to do" — internal tools, BI dashboards, CRM data entry, almost everything an analyst opens during the day).
The downstream consequence is the one that matters for anyone building software: new line-of-business tools are increasingly designed assistant-first. The UI exists for the cases the assistant can't handle alone, or for the moments where direct manipulation is genuinely faster than describing what you want. That's the shift that makes the analogy bite — not that users adopt assistants, but that the next generation of software assumes them.
Why this matters for regulated industries specifically
We work mostly with regulated financial and professional services firms in NZ and AU. The temptation in regulated industries is to assume this transition will arrive later for you, because it has to clear compliance, security review, change advisory board, vendor risk, the lot. That instinct is half right and half dangerous.
Half right: yes, the adoption curve in your sector will be slower, and the controls around how an assistant is allowed to act on customer data are genuinely harder problems than they are in a consumer app. Don't let anyone tell you otherwise. The change-management problem here is real and it's the thing we spend most of our time on.
Half dangerous: the underlying technology shift is not waiting for your industry to be ready. The control plane is moving regardless. Firms that treat AI as "a tool we'll evaluate next budget cycle" are making the same bet as the IT shop in 2002 that treated SaaS as a passing fad and stuck with on-prem everything. They weren't wrong for a few years. They were materially behind by the end of the decade, running an operating model their competitors had quietly outgrown.
The firms that came out ahead in the browser transition weren't the ones who picked the right browser. They were the ones who understood, early, that the browser was where the control plane was going to live — and rebuilt their operations around that assumption while their competitors were still buying desktop licences.
What we're not saying
A few things this post is deliberately not claiming:
- Not zero-sum. Plenty of work will stay in dedicated apps, on the desktop, on phones, on paper. The browser didn't kill the desktop; it just absorbed the centre of gravity. AI will do the same.
- Not a foregone conclusion. A regulatory event, a major safety incident, a technical wall on capability, an antitrust intervention — any of these could re-order the deck. Anyone telling you the next decade is settled is selling something.
- Not a Claude-only future. We use Claude because we think it's the best tool for our clients today, in our market, for the work we do. That's a judgement that gets re-made. The point of the thesis is that some leading assistant occupies this position — which specific one, and how many credible alternatives sit alongside it, is a separate question we're not trying to answer here.
- Not "implementation is easy now." This is the part of the analogy that gets people in trouble. The browser made distribution easy and implementation hard in a different way — Salesforce didn't win because SaaS was easy, it won because someone did the hard work of turning a category into a product. The same will be true here. Picking an assistant is the easy part. The work is everything around it: the integrations, the permissions model, the audit trails, and — hardest of all — the change inside the organisation.
The honest version of the headline
So: is Claude the new browser?
The honest version of the claim is that one of the leading AI assistants is going to occupy roughly the position in the next decade that the web browser occupied in the last two. We think Claude is the best-positioned of the current crop to be that assistant for serious work in regulated industries — partly because of the underlying model, partly because of how Anthropic has approached safety, partly because the ecosystem (MCP, the API, the IDE integrations) is starting to compose in the way the web stack composed in the early 2000s.
We could be wrong about which one wins. We're not wrong about the shape of what's coming.
If you're running a financial services firm in NZ or Australia and you've been treating this as a "watch and wait" question, the calendar matters more than the answer. The firms that moved early on web didn't do it because they were sure. They did it because they understood the cost of being late was higher than the cost of being early to the wrong specific tool.
That's roughly where we think your industry is, right now, on this one.