[AI SPRINT] The Tools Are Moving Fast. Are You?

Every few months, a wave of AI releases hits that feels less like routine product news and more like a reset in where the market is heading. This is one of those weeks.

In just the past week, OpenAI launched GPT-5.3 Instant for everyday AI improvement, then GPT-5.4 with higher reasoning quality gains, added reusable Skills, and brought ChatGPT directly into Excel. Anthropic expanded Claude further into daily productivity, with Claude now available in both Excel and PowerPoint, backed by a growing marketplace of industry-specific skills. And then, this morning, Microsoft announced something that redraws the competitive map: Copilot Cowork, built in direct partnership with Anthropic, with Claude now available across all of Copilot, not just in select features.

Taken together, these launches point to the same shift. The primary AI platforms people are already building around are extending their capabilities into the software where work actually happens. And one company is showing up at the center of that shift more than any other.

Microsoft just made a very public bet. And it wasn't on OpenAI.

GPT-5.3 and 5.4: What Actually Changed

OpenAI now has two models doing most of the heavy lifting in ChatGPT, and it's worth knowing the difference before you default to whatever's selected.

GPT-5.3 Instant is the default model for most logged-in users. It's fast, reliable, and built for routine work. For everyday drafting, summarizing, brainstorming, and quick analysis, it will handle most of what you throw at it.

GPT-5.4 Thinking is where the more meaningful gains show up. It's built for harder tasks and produces stronger results when the work has more structure, more reasoning, or more room for error. It's 33% less likely to make errors in individual claims compared to its predecessor, and overall responses are 18% less likely to contain errors. For business outputs like client documents, proposals, and research summaries, that quality gap matters. Less cleanup is not a small thing. It's the difference between something being useful and something being ready.

OpenAI also focused 5.4 heavily on productivity work, especially spreadsheets and presentations. On an internal benchmark of spreadsheet modeling tasks, GPT-5.4 scores 87.3% versus 68.4% for the prior version. For presentations, human raters preferred GPT-5.4 outputs 68% of the time, citing stronger aesthetics and better use of image generation. You don't need the Excel add-in to see this. Ask GPT-5.4 Thinking to build a presentation directly in ChatGPT and you'll notice the difference right away.

To access it: open the model picker in ChatGPT and select GPT-5.4 Thinking. Many power users are now running it as their standard model.

There's also a workflow change worth noting. When GPT-5.4 Thinking takes on a complex task, it shows its plan upfront. You can guide it while it's still working instead of waiting until the end, spotting what it missed, and starting over. That may sound minor, but it makes the model much easier to use on real work.

Skills are also rolling out in ChatGPT. This concept was pioneered by Anthropic in Claude and is quickly becoming the direction the whole industry is moving. Think of Skills as reusable playbooks. Instead of re-explaining how you want a weekly status report, proposal draft, or research brief handled every time, you save the workflow and reuse it. For teams, that means more consistency and far less prompt babysitting. It may end up being one of the highest-leverage features most businesses still aren't using.

How to get started: access Skills under your profile icon, then Skills. Enterprise and Edu admins need to enable the feature in workspace settings first. We'll do a full breakdown on building and applying Skills in an upcoming issue, but start exploring now.

Your AI Platform Is Showing Up Where Work Happens

The biggest productivity story right now isn't just better models. It's that the AI platforms people are already building around are extending into the software where work actually gets done.

That distinction matters.

For a lot of teams, ChatGPT, Claude, or Copilot has become a primary AI environment. It's where preferences get set, workflows get built, and trust gets established. What changed this week is that those same capabilities are now reaching further into Excel, PowerPoint, and other work tools, so users can bring the same intelligence into the flow of work without constantly switching contexts.

OpenAI's new ChatGPT for Excel beta is one example. Claude in Excel and PowerPoint is another. Microsoft's Copilot Cowork pushes in the same direction at the enterprise level. In each case, the pattern is the same: the AI platform stays central, but its usefulness grows because it can now operate closer to the files, systems, and tasks people are already working in.

ChatGPT for Excel brings ChatGPT directly into the workbook so users can build models, run scenarios, and generate outputs without bouncing back and forth to a separate chat tab. It's powered by GPT-5.4 and rolling out now in beta.

Claude in Excel goes further. It reads complex multi-tab workbooks, explains calculations with cell-level citations, and safely updates assumptions while preserving formula dependencies. It also connects to real-time data feeds from partners including Moody's and LSEG, so you can pull live market data directly into your models.

Claude in PowerPoint extends the same idea into presentation work. It reads your existing deck including layouts, fonts, colors, and slide masters, then generates or revises slides that stay aligned with the structure and style already in place. The painful part of deck work is rarely the first draft. It's the endless cleanup and iteration after that. Claude handles more of that loop.

Put those moves together and the direction is obvious. The major AI platforms are not just trying to win the chat experience. They are trying to make their capabilities available wherever the work itself is happening. Once users get used to bringing the same AI system into the tools they already use, the platform relationship gets much harder to walk away from.

How to install:

• ChatGPT for Excel (beta): Sign in with your OpenAI account. Available to Plus, Pro, Team, Business, Enterprise, and Edu users in the US, Canada, and Australia. Note for Business and Enterprise users: it's off by default and an admin must enable it.

• Claude in Excel: Sign in with your Claude account. Available on Pro, Max, Team, and Enterprise plans.

• Claude in PowerPoint (beta): Same Claude account sign-in. Available on Pro, Max, Team, and Enterprise. Usage limits are doubled through March 19, 2026.

All three install through the Add-ins menu inside Excel or PowerPoint. No IT department required for individual users.

Skills Are Becoming the Real Product

Another important thread this week is the rise of reusable workflows, and this is where things start to get more interesting.

The value is no longer just access to a powerful model. It's increasingly in packaging that model into workflows that reflect how real professionals already work. Not "ask me anything," but "run this recurring process the way my team actually does it." That's the leap from novelty to leverage.

OpenAI is rolling out Skills in ChatGPT. Anthropic is already further along, with a marketplace of prebuilt skills aimed at investment banking, equity research, private equity, wealth management, HR, design, engineering, and operations.

Financial services is the clearest example of where this is heading. Anthropic isn't just offering a model to answer questions. It's packaging workflows around client meeting prep, financial plan building, portfolio rebalancing, document review, model updates, and research drafting. Partner plugins from LSEG and S&P Global bring live financial data directly into those workflows, so Claude isn't just running analysis. It's pulling live yield curves, equity prices, and market data as it works. Add the ability for enterprises to build private plugin marketplaces encoding your firm's specific processes and templates, and this starts to look a lot less like a chatbot and a lot more like an operating system for work.

That's where a lot of the real competitive advantage will come from next.

Microsoft Bets on Anthropic

This morning, Microsoft announced Copilot Cowork. And the competitive landscape shifted in real time.

Working closely with Anthropic, Microsoft brought the technology that powers Claude Cowork directly into Microsoft 365 Copilot. Copilot Cowork lets you delegate complex, multi-step tasks to Copilot. It breaks the work into steps, reasons across your tools and files, and carries it forward with visible progress and opportunities to steer. Tasks can run for minutes or hours, coordinating actions across Outlook, Teams, Excel, and PowerPoint and producing real outputs along the way.

The practical example Microsoft demonstrated: preparing for a customer meeting by building the presentation, assembling the financials, emailing the team, and coordinating calendar time, all from a single instruction, all while keeping you informed and in control.

There's something larger happening here too. Microsoft initially built all of Copilot around OpenAI's models. That's no longer the case. Claude is now available across mainline Copilot Chat for Frontier (beta) program users, sitting alongside OpenAI models as a general-purpose option in everyday Copilot conversations. This follows a $30 billion Azure compute deal with Anthropic in November 2025, the integration of Claude Opus 4.6 into Microsoft Foundry in February 2026, and widespread internal adoption of Claude Code across Microsoft's own engineering teams.

Microsoft didn't just partner with Anthropic on a feature. They built their flagship agentic product on top of Claude's technology.

For businesses already deep in Microsoft 365, Copilot Cowork may be the most natural path. For others, standalone Claude Cowork may still offer more flexibility. The two tools aren't identical. Claude Cowork gives you more flexibility and access to Anthropic's full plugin and skills ecosystem. Copilot Cowork gives you that same agentic engine running inside your existing M365 data and security framework. For most businesses, the right answer depends on how deeply you're already embedded in Microsoft 365 and whether you need enterprise governance or individual flexibility first.

Copilot Cowork is currently in research preview with a limited set of customers. Broader access is coming through Microsoft's Frontier program in late March.

The Microsoft Pricing Conversation Just Got More Complicated

Here's a tension worth naming directly. Microsoft is raising prices at the same time it's announcing it needs Anthropic's technology to compete.

The new Microsoft 365 E7 bundle is was also announced today at $99 per user per month, compared with $60 for the current E5 subscription. That's a 65% price increase. E7 bundles Microsoft 365 Copilot, the new Agent platform, plus previous tools.

The case for E7 is legitimate for larger organizations if they can actually get their teams trained to use it. It consolidates tools you'd otherwise buy separately and adds governance for AI agents that enterprise IT teams genuinely need.

But here's what matters for SMB and mid-market leaders: Microsoft is charging 65% more for a productivity suite that now prominently features a third-party AI you can access directly for $20 per month. And on top of that, few organizations are finding success with Copilot itself, and now are being told their upgrade path to get value is now significantly more than best of breed competitors.

Before your next Microsoft renewal, ask three questions:

1. What is Copilot actually delivering to your team today, not in theory, but in daily practice?

2. Does your team need the enterprise governance and M365 integration that Copilot Cowork provides, or would standalone Claude Cowork cover your needs at a fraction of the price?

3. Do you have the ability to train your workforce on all the tools of E7 so you can justify $1200 per year for them? Adoption is the biggest challenge today, so purchasing E7 with the hope they use it will be a money-loser.

That conversation is better to have now than after May 1.

One More Worth Watching: Perplexity

While the major platforms race to extend themselves into more of the software stack, Perplexity is carving out a different role.

Its strength isn't replacing your main AI platform. It's becoming a specialist for research-heavy work and adding similar “cowork” agent functionality. Perplexity Computer executes complex research workflows using 19 different AI models, creating subagents to handle specific problems within a larger task. It's built for collecting data across multiple sources, running legal or financial analysis, and delivering finished outputs rather than just answers. The company also moved away from advertising entirely, saying it undermined users' trust in the accuracy of results.

If your team does intensive research work like competitive analysis, market research, or due diligence, Perplexity Pro deserves a spot in your toolkit. Not as a replacement for your primary AI platform. As a specialist. The market may not settle around a single winner. It may settle around a small number of primary platforms plus a few specialists that earn a permanent place because they're exceptionally good at one thing. Perplexity increasingly looks like one of those.

What This Means for Your Business Right Now

The pattern across all of these updates is pretty clear. The AI market is maturing around a smaller number of serious platforms, and those platforms are expanding their reach. Model quality still matters, a lot, but the battle is increasingly about more than the model itself. It's about whether your primary AI platform can show up in more of the places where work already happens, carry your workflows with it, and become more useful as a result.

The businesses that get the most value from AI won't be the ones that subscribe to the most tools. They'll be the ones that choose a primary AI platform well, then take advantage of the growing set of ways it can support work across the rest of their software environment.

Three things to do this week:

1. Use GPT-5.4 Thinking on something real. Not a toy prompt. A proposal, research summary, client deliverable, or strategic document. See whether the quality difference actually changes the amount of editing you need to do afterward.

2. Test one of these integrations in a real workflow. Use ChatGPT in Excel, Claude in Excel, or Claude in PowerPoint on a file you already work with. The point isn't novelty. It's to see what changes when your primary AI platform is available inside the software you already use.

3. Identify one repeatable workflow that should become a Skill. Weekly reports, proposal drafts, meeting prep, account planning, research briefs, and internal summaries are all good candidates. That's where consistency starts to compound.

The important question is no longer just which model is best. It's which AI platform is becoming most useful across the way your team already works.

Trent Gillespie is CEO of Stellis AI and a keynote speaker helping business leaders understand and operationalize AI in their companies. He spent almost nine years leading global innovation efforts at Amazon before leaving to help other companies build the capabilities they need to compete. Book Trent to speak to your group or book a call to discuss using AI within your business.

Did someone forward this newsletter to you? If you're not already signed up, you can subscribe to AI SPRINT™ for free here.