Nate B Jones demonstrates Claude AI building financial models and board decks inside Excel and PowerPoint in minutes
A solo presentation by Nate B Jones on how Anthropic's Claude AI, integrated into Microsoft Excel and PowerPoint, can perform a full day of analyst-level financial work in under an hour.
Summary
Nate B Jones demonstrates that Claude (running on the Opus 4.6 model) can build a complete operating model in Excel and a full board presentation deck in PowerPoint in roughly 30 minutes total — work he says a Goldman Sachs analyst confirmed would typically take a full day. He argues this is not a niche coding or research story but a transformation of the tools that over a billion people use daily, and that the shift is already in production at institutions including Goldman Sachs, AIG, and Norway's sovereign wealth fund. His central argument is that the application layer — Excel, PowerPoint — is becoming a "dumb pipe," with real value migrating to the AI intelligence layer that Anthropic controls. He closes with a pointed warning: the collapse in the cost of producing professional artifacts will flood organizations with polished but hollow "work slop," and the professionals who survive will be those with the judgment to know what is worth making — not just the ability to make it faster.
Key Takeaways
FULL TRANSCRIPT
What shipped and why it matters more than the benchmark headlines
Nate B Jones: I used Opus 4.6 to build in ten minutes what it takes a Goldman analyst to build in a day. I'm not a Goldman analyst. And then I built a board deck for that in just twenty more minutes. General intelligence just showed up in Excel and PowerPoint, and its name is Claude.
I built a full operating model last week — revenue projections, cost structure, unit economics, the works — and that took just a few minutes. After that, Claude in PowerPoint took over and built slides, executive summaries, financials, and key metrics using an actual slide deck template. Just a few minutes later, I had a presentation with charts referencing live Excel data, formatted in the correct fonts and colors. This is something that would have taken a couple of days just a few months ago.
And yes, I did mean it about Goldman. A Goldman Sachs analyst looked at the model and told me it was solid — the kind of output that would have probably taken him a day to build — and it took me thirty minutes total, with a deck included.
This is the piece of this week's news that most people are going to sleep on, because it doesn't have the drama of a benchmark score. It doesn't have the spectacle of sixteen agents building a compiler, or an agent managing fifty developers. This is just Excel. It's just PowerPoint. The tools nobody thinks about twice. And yet we spend our days there. And as of this week, they effectively have general intelligence inside them — the same intelligence that built that C compiler, the same intelligence that found five hundred zero-day vulnerabilities on its own after security researchers had passed the code as secure.
Here's the point that should really stop you. It's not about a single product release. I don't care if you think Opus 4.6 is the best model or not. The point is that this shift into Excel and PowerPoint paves the way for 4.7 and then 5.0. The applications are not going to change — PowerPoint will look the same, Excel will look the same — but the intelligence inside them is going to compound. This is the dumbest Excel and PowerPoint will ever be.
I covered Opus 4.6 in a separate piece earlier this week — what the model can do, why it matters, agents. This piece is about what happens when that intelligence shows up in tools that over a billion people use every single day, and why it turns Microsoft into just a dumb pipe, and what that means for how you think about work when your tools are getting smarter faster than you can update your assumptions about them.
What actually shipped and how to get it
So what actually shipped? Two things happened in the past couple of weeks, and taken together they represent something bigger than either one of them individually.
On January 24th, Anthropic opened Claude in Excel to Pro subscribers — anyone paying twenty dollars a month or more. The feature had been in limited beta since October of last year, but the January release made it broadly available. Then on February 5th, alongside the Opus 4.6 launch, two things happened at once: Claude in Excel upgraded to 4.6 — the same model that powered all of those remarkable coding results — and Claude in PowerPoint launched for the first time.
The Excel integration is not a chatbot bolted onto the sidebar, even though it looks like it. It actually operates directly against your work. It reads your existing data, understands your tab structures, writes and debugs formulas, builds pivot tables. Yes, it is absolutely not perfect. Yes, it needs work sometimes and a check from an experienced analyst. But I'm going to keep reminding you: this is the dumbest that model is ever going to get in Excel.
The PowerPoint integration is more interesting than most coverage suggests. It doesn't just generate slides. It reads your slide masters, your layouts, your fonts, your color schemes, your template, your font hierarchy. It produces slides that don't look like AI made them because they match the design system your team already uses. That has been a huge breakthrough for AI in the past month or two that most people have slept on. Back in the fall of 2025, building an AI PowerPoint meant you had to give up your own templates. Not anymore.
The combination of Excel and PowerPoint together matters more than any of these tools individually. Because both run on the same underlying model, Claude is able to bring the same intelligence to bear across both. Claude-produced Excel documents play very nicely with Claude-produced PowerPoints. So if you're building an analysis in Excel and then you tell Claude in PowerPoint to generate the board deck off of that analysis, it's going to be very easy to get from data to decision in a single sitting. That's the promise of working with Claude seamlessly across a range of different Microsoft artifacts.
Now, the part most coverage skips — how do you actually get it? Claude in Excel is available now to everyone on Claude's Pro plan at twenty dollars a month. That's it. Same price as Netflix. Just install the Claude desktop app, enable the Excel integration, and it appears inside Excel.
Claude in PowerPoint is harder to get right now. It launched on February 5th and is currently only available on the Max plan, which is around a hundred dollars a month. I think it's because you burn more tokens on PowerPoint than Excel. It's not yet out on Pro. If you need both tools as an individual, the Max plan is the only option you've got.
The pricing matters because of what it implies about the cost of intelligence. Junior financial analysts can cost six figures — a hundred thousand, a hundred and twenty thousand dollars fully loaded. An associate at a consulting firm can bill three hundred to five hundred dollars an hour. At twenty to a hundred dollars a month for Claude in Excel and PowerPoint, a lot of organizations are going to start asking themselves whether junior analysts are adding incremental value. I'm not saying the junior analyst role is obsolete. I'm saying the junior analyst who only builds models and decks manually has a very big problem, because that scarce skill is no longer scarce. You're going to start to get measured by how quickly you can ramp on AI tooling and expand your scope.
Financial data connectors and pre-built workflows
Here's where it gets really interesting, because most people comparing Claude in Excel to Microsoft Copilot miss the point entirely. They're stuck talking about formulas and native integrations.
Anthropic partnered with Moody's, the London Stock Exchange Group, Thirdbridge, and others to build financial data connectors directly into the Claude ecosystem. These are not generic web scrapers. They are authenticated, structured data feeds from platforms that institutional finance runs on. In practice, you can ask Claude to build a comparable company analysis, and instead of manually pulling data from a terminal, the model will query live financial data through those connectors and populate your spreadsheet with real numbers.
Anthropic also shipped pre-built financial skills — purpose-built workflows for the tasks that eat most of an analyst's week: comparable company analysis, discounted cash flow models, due diligence data packs, and so on. These aren't templates that you fill in. They're intelligent workflows that understand what a discounted cash flow model actually needs and how to structure the assumptions tab to support it.
For anyone who has built a DCF model from scratch, you know the mechanical work involved takes a long time — not because it's conceptually difficult, but because there are hundreds of cells that all need to reference each other correctly. Anthropic identified that pile of mechanical work and realized that with the right data feeds and the intelligence of Claude, they could knock it out and make the spreadsheet dumb plumbing. The intelligence is what matters.
Is this real or is it a demo?
I need to address the "is this real or is it a demo?" question head-on, because I get so many questions in the comments after videos like this.
Anthropic announced a Goldman Sachs partnership on February 6th, and that partnership had already been ongoing for months while Goldman essentially pioneered this behind the scenes. Goldman is deploying Claude across accounting and compliance workflows now as a production tool. When the most prestigious investment bank in the world puts this in production for internal operations, the demo question gets answered.
AIG reported that Claude made their document reviews five times faster, with accuracy improving from 75% to over 90%. Not faster at the expense of quality — faster and more accurate at the same time. The error rate went down while the speed went up. That was impossible in the 2010s era of software. It is a signature of a tool that does not get tired, that does not skip the boring rows, that does not assume the numbers in the summary tab match without checking.
Meanwhile, the bank that manages Norway's $1.7 trillion sovereign wealth fund reported an estimated 213,000 hours saved from Claude in Excel. This is what it looks like when you target a pain point that is scaled across a one-and-a-half-billion-user base in Microsoft. And this is why Microsoft should be worried — because Claude's entire strategy disintermediates Microsoft's influence on their own user base, inside their own tool.
Specific workflows across finance and beyond
Let me walk through a few specific workflows, because not everybody is a financial analyst and I want to give you a sense of what general intelligence in your tools actually looks like.
Let's start with an operating model. You open a blank workbook, you have a dream of a small business, and you tell Claude: "Please build me a three-year operating model. Here is my revenue target, here is how many people I want to hire, here is how many customers I want to get, here is the kind of product I have and what I want to sell it for. I don't know how to build a business operating plan. I need your help." It turns out Claude does a really good job at that. It may not be perfect, but it gets you about ninety to ninety-five percent of the way there out of the gate.
How about a board deck? Let's say you've built the model and you want Claude to build a PowerPoint you can show your banker to get a small business loan. Claude can read the Excel file you upload, understand it — because again, it's the same intelligence underneath both — generate charts that reference actual numbers, apply your company's slide template, and put out something you can go to a banker or an investor with.
Let's say you're a startup founder pitching a Series B. You have your financials in Excel and a pitch template your designer built last quarter. All you have to do is tell Claude: "Build a twelve-slide pitch deck from these financials. Here's where we want to go. Here's the arc of the story." Claude will do it. It will use your fonts, your colors, your layout grid. And by the way, that is all new since last fall. The last time I talked about Excel, I had to say: "It's amazing, it does PowerPoint, but one — it's not actually in PowerPoint, and two — good luck using your own templates and layouts." Not anymore. That's how fast things move.
What about due diligence? Let's say you're trying to understand a small business you want to buy. You upload three years of financial statements and tell Claude: "Please build me a due diligence data pack and flag anything unusual." Claude saves you dozens of hours combing through those statements and greatly increases the probability you'll catch a red flag that might stop the acquisition and save you from a bad deal.
Let's say you're a product manager who wants to do a comparable company or comparable product analysis. You name a few competitors and Claude can pull all of the relevant trading data, plus the product data, and build a competitor spreadsheet analysis from scratch in just a few minutes.
What about a quarterly business review? Your department heads submit their numbers in separate spreadsheets. You consolidate it all in Excel. You tell Claude in PowerPoint: "Please build me the QBR deck with fifteen slides using our corporate template and these numbers." Done.
Now, all of these are finance workflows, but there are non-finance workflows that are just as important. What about a strategy analysis? You have a spreadsheet of fifty competitors with market positioning and recent funding rounds. You want to score each competitor on six different dimensions, weighted by strategic priorities. You give it to Claude in Excel. Then you give it to Claude in PowerPoint and it builds a competitive landscape deck with a positioning matrix, a threat assessment by segment, and recommended strategic responses.
What about sales enablement? Your sales team sends the same ten-slide pitch to every prospect. Why not hand Claude the company's CRM data and their last three earnings transcripts and tell it to customize the pitch for a CFO audience at a mid-market manufacturing company? That is trivial to do.
What about HR and people analytics? You export twelve months of employee survey data — two thousand responses, free-form text, Likert scales, eight departments. You tell Claude in Excel: "Analyze the sentiment by department, identify the three strongest predictors of attrition risk, and build a summary dashboard." It'll do it.
What about program management? You have a master tracker in Excel with two hundred line items across a dozen work streams — owners, deadlines, status, dependencies. Claude can produce a program status deck for the steering committee in one shot.
And before you even get to the headline workflows, there's the daily grind that Claude in Excel can eliminate: debugging a VLOOKUP chain that breaks when someone sorts a column, writing a Power Query transformation to clean vendor data, building conditional formatting rules, tracing a circular reference across four tabs. This is the kind of work that eats hours a day when you live in spreadsheets. It's the first thing Claude handles, and the thing that frees up the most time before you even start the big workflows.
Every one of these workflows exists today — not next quarter, not on a waitlist. Right now.
The context layer and why the time savings multiply
Here's what none of the individual workflows make obvious enough: the time savings don't just add up, they multiply.
Having Claude in Excel saves you time on modeling. Having Claude in PowerPoint saves you time on decks. Having both saves you more than twice the time of either alone, because it eliminates an entire category of work that existed solely because the tools could not understand what each other built in the age before shared intelligence.
Think about what actually eats your week. It's not just building the model. It's not just building the deck. It's the mental work that comes from the translation layer in between. You finish the analysis in Excel, then you open PowerPoint, and you have to start re-explaining the same data. You have to think about how it changes when you position it in deck form versus spreadsheet form. That translation cost is where most knowledge workers spend the majority of their productive hours. It's not even thinking — it's translating into a different format for a different audience.
When one intelligence spans both tools, that translation cost starts to drop toward zero. Claude doesn't just export data from Excel and import it to PowerPoint. It deeply understands the data in Excel because the same intelligence built it, and it carries that understanding directly into the presentation. The chart it builds in PowerPoint reflects an understanding of what the model extracted from the analysis. The narrative on the slide reflects a deeper interpretation that the model formed when building the Excel spreadsheet. Context flows more easily because the same model is building both.
I'm not saying there's a direct export from Excel to PowerPoint today, but I would bet you a lunch that is coming in the next couple of months. And in the meantime, having a model that understands how both Excel and PowerPoint work and can easily translate context between them is a godsend.
What we're talking about here is the context layer, and it is the future of work. It's not really about the application layer anymore — Microsoft owns that. It's not even necessarily about the data layer — your databases own that. The context layer sits between them. It's the AI's accumulated understanding of your data, your brand, your audience, your goals. Every time the model touches a new tool, the context layer gets a little bit richer. Every time it sees how your board deck differs from your team Slack update, it learns something about how your organization translates information for different audiences.
Applications are containers. Data is raw material. The context layer is what Anthropic is making a play for here — the intelligence that understands what the data means and how to express it for different audiences in different formats. That is where the value is accumulating, and that is what Anthropic is laser-focused on with Claude in Excel and Claude in PowerPoint. And unlike the application layer, which Microsoft has owned for decades and which barely changes year after year, the context layer improves automatically with every single model upgrade and every new tool integration. It's the fastest-compounding asset in your tech stack, and most organizations don't even know it exists.
The upgrade cycle that changes everything
That's what separates what happened this week from a normal product launch. On Tuesday night — the night before the Opus 4.6 launch — Claude in Excel ran on Opus 4.5, a strong and capable model. On Wednesday morning, it ran on Opus 4.6. Nobody installed anything. Nobody downloaded a patch. Nobody sat through a migration wizard. The spreadsheet looked the same, but it suddenly had dramatically more context, better reasoning, and the ability to hold an entire multi-tab model in working memory and understand how every cell relates to every other cell.
Think about what that means for the next upgrade. Opus 4.7 is coming. So is 5.0. Each time a new model ships, every Claude-powered Excel and PowerPoint on Earth gets smarter overnight without you doing anything. The operating model that took ten minutes with Opus 4.6 might take five with 4.7 and be ninety-nine percent right instead of ninety-five. The pitch deck that needed twenty minutes of back and forth with 4.6 might need five minutes with 5.0. The quality of reasoning continues to improve. The depth of analysis deepens. The output moves closer to perfect. And it's not because you learned a new tool — it's because the tool learned on its own and got better.
This is a fundamentally different upgrade cycle from anything the software industry has produced. Microsoft ships a new version of Office every few years. Feature updates land quarterly. The pace of improvement is set by the software company's release schedule, its engineering priorities, its QA cycle. With Claude, the pace of improvement is set by Anthropic's model release cadence — and those are happening every couple of months with capability jumps that would be measured in years by traditional software standards. That three-month gap between 4.5 and 4.6 saw context expand fivefold. What is 4.7 going to bring?
Almost certainly, your mental model of what AI tools can do is now behind reality. It is hard to keep up with how fast reality is moving right now. The task that Claude maybe couldn't handle last month in Excel — maybe it handles it now. The presentation quality that wasn't sufficient in January because it didn't match your templates — maybe it works now. And by April, both will have improved again.
The assumption that you learn your tools once and they stay the same — that the thing you tested last quarter is the same thing running today — that assumption is dead. Your tools are getting smarter faster than you're updating your expectations of them. The practical consequence is that you need to re-evaluate your workflows continuously. Not annually, not when someone sends you a blog post. All the time. Because the boundary between the tasks you do yourself and the tasks it makes sense to delegate to AI just keeps moving, and it's moving in one direction, and it's moving fast.
Claude versus Microsoft Copilot — and why that's the wrong comparison
I can hear the Microsofties in the comments saying, "Doesn't Microsoft Copilot already do this?" The answer is sort of, and the real answer leads somewhere more important than a feature comparison.
Copilot's advantage is native integration. It's built into Microsoft 365 from the ground up. The UI is seamless. It understands the Office ecosystem in a way a third-party tool doesn't. If your organization lives within Microsoft, Copilot is the path of least resistance, and sometimes it's sold that way.
Claude's advantages are reasoning depth, local file support, and financial data connectors. Claude wins on tasks that require genuine reasoning over complex, multi-step problems — like debugging a formula chain across twelve tabs, or structuring an analysis that requires judgment about what matters. The local file setup matters as well. Copilot will require OneDrive for most of its functionality, which means your files live in the Microsoft cloud. For organizations handling sensitive financial data, it's useful to have an alternative.
But in the end, the Copilot comparison is the wrong frame for what's actually going on. The real story is more structural. In September 2025, Microsoft added Claude models to its own Copilot. Yes — the company that invested thirteen billion dollars in OpenAI, and built Copilot on OpenAI's models originally, hedged by putting a competitor's intelligence inside its own product. When the company that owns the application layer starts offering someone else's intelligence inside it, that tells you where the value is migrating.
Microsoft really is becoming a dumb pipe. Not overnight, not completely, but the pattern is unmistakable. It mirrors what has happened to every platform caught between a commoditizing interface and a rapidly improving capability layer. AT&T built the network, and then the network became a pipe for Google and Netflix. The value migrated from the carrier to the service. Browsers were supposed to be the platform, and then they became rendering engines for web applications. Value migrated from the container to what ran inside it.
Excel is a grid of cells. It has been essentially the same for twenty years. New features are at the margins. PowerPoint is just a canvas for slides — same story. The intelligence layer is what is compounding. The application layer is frozen. And that is why Microsoft is hedging by offering every major AI model inside its own products. That's exactly what a dumb pipe does — it carries whatever traffic flows through it.
The implication for your organization is to stop thinking quite so much about your tool choice and start thinking about your intelligence choice. The question is not "should we use Excel or Google Sheets?" It's "which AI model powers our spreadsheet, and is it the best one for the work we do?" You need to start thinking of applications as containers and asking where the intelligence is coming from and whether it has the value you're looking for.
What happens when artifacts go to zero
This is the thing we aren't talking about enough. If the cost of producing these artifacts is collapsing toward zero extremely rapidly, what happens when these artifacts are free?
So much of our traditional human-driven professional services model starts to break apart. Consulting breaks — not because consultants are unnecessary, but because the business model depends on a very large time component that is about to disappear. When a deliverable that used to require forty hours of associate time can be produced in forty minutes, that whole model isn't going to work anymore.
The correct response is not panic. It's recognizing what becomes valuable when artifacts go to zero.
Analysis is becoming a commodity. Judgment is becoming very, very valuable. Knowing how to build a discounted cash flow sheet — Claude can do that. Knowing which assumptions you want to stress test, which scenarios to run, what stories the numbers are telling, and how to read the Claude spreadsheet and find the mistakes — there's judgment there, and judgment is what clients are going to pay for and boards are going to need.
The people who thrive in this environment are not the ones who build the best artifacts with AI. They're the ones who know which questions to ask before whatever the model is building gets built. They're the ones who can look at a completed analysis and say, "This is technically right, but the whole question is wrong — it's framed wrong." That's value. That's where human value is going. They're the ones who understand the business well enough to know which of the seventeen possible analyses Claude produced is the one that should actually drive the decision.
This is the strategic skill I keep coming back to. When production is free, economic returns flow to people who know what's worth making — not necessarily more of, not necessarily better, not even faster. Because the ten-minute operating model is worthless if you're modeling the wrong thing. The thirty-minute board deck is worthless if it tells a story that doesn't match reality and doesn't line up with investor expectations. The tool will make you faster. Only you can make sure it's right.
The wave of work slop
There's an uncomfortable truth hiding inside all of this capability, and that is a tidal wave of slop. Every tool that makes it easy to produce excellent work makes it just as easy to produce garbage. We are about to drown in AI-generated content that looks professional. Researchers have started calling it "work slop" — AI-generated professional content that looks technically competent and is completely hollow. The estimated productivity cost is $186 per employee per month in time wasted processing it. That sounds like it means something and says nothing. It adds up, and I bet that figure is understated.
This isn't really about an AI adoption problem. It's not about whether you rethink your workflow or just bolt AI onto the old one. This is a fundamentally different problem. It's about whether you have the judgment to know which work should exist, and whether you work on a team that displays that judgment as well.
The same capability that lets a thoughtful strategist produce a day's work in ten minutes lets a careless operator produce a week's worth of polished nothing in an afternoon. The tool will never know the difference, because you are the one who maps what is needed onto the business context and what the market requires. That's on you.
In this sense, taste — which gets talked about a lot — is not an aesthetic preference, and it's not something impossible to learn. It's simply the ability to distinguish between output that serves a meaningful human purpose and output that just exists. It's knowing that a forty-slide deck Opus can create may look impressive, but it's not as valuable as a ten-slide deck. It's knowing the third scenario in the model is what the board needs to see, not the other two. It's knowing when the analysis is done and you shouldn't add more data because that's just going to dilute things.
Organizations that have people with good judgment are about to massively outexecute organizations with the same tools in the same industry that don't. Good judgment is about to supercharge economic activity for organizations that understand how to deploy it.
The end of the execution premium
For thirty years or more, professional value has been built on execution skills. Can you build the model? Can you write the code? Can you design the spreadsheet? Can you structure the analysis? Those execution skills created our whole modern knowledge economy. They're what universities teach. They're what hiring managers were taught to screen for.
That execution premium is evaporating now. Not in five years. Now. The ten-minute operating model isn't a preview of a future — it's a product you can buy today.
But what is not evaporating is the thinking that sits above the execution layer. We have to move up a level of abstraction in our work as knowledge workers — all of us. It's now the ability to frame the right question. It's the strategic awareness to know which analysis matters. Because the tools are going to keep getting better, the thinking is the place that has to hold the value.
Claude can build the vehicle for your thinking, but Claude cannot replace human judgment. And in that sense, I think Anthropic has done a great job framing Claude as a tool for human thinking — similar to a chalkboard or a notebook. That's the right frame, and it's a very compelling one for professionals who are looking to elevate their work in the age of AI.
We have to do better, because AI is coming for the traditional execution skills that defined knowledge work. The models are going to keep getting better. What was ninety-five percent good will be solved by the middle of the year. It's your ability to say "this is the right direction to go in" that is going to make or break your career in 2026 and 2027.