AI Just Got Serious About Small Business. Here's What Actually Changed.

In one stretch of May, a few things happened that, taken together, mean small business AI just stopped being a science experiment.

Anthropic launched a dedicated Claude for Small Business offering. KPMG announced it’s rolling Claude out to its workforce of more than 276,000 people. PwC put tens of thousands of its professionals on Claude. Anthropic also acquired Stainless, the company behind the SDK tooling — software development kits, the libraries developers use to wire AI into their own software — that powered much of the industry, reportedly for over $300 million. Microsoft kept shipping Copilot updates. Google announced Gemini 3.5 Flash and a new multimodal model called Gemini Omni at I/O.

Any one of those would be noise. All of them inside two weeks is a pattern. The big AI vendors have stopped pretending small business is a side project. They’re building products, pricing tiers, and integration partnerships aimed at companies that look like yours.

The question for a business owner is no longer “should we be paying attention to AI.” It’s “given how fast this is moving, what should we actually do with our money and our time this quarter.”

What “Claude for Small Business” actually is

This is the headline most likely to matter for you. Anthropic launched a packaged offering explicitly named for small businesses. Until now, the choices for a 30-person company were:

• Use the consumer version of Claude.ai or ChatGPT and accept the data governance risks (we wrote about the data leakage problem here)
• Pay for Microsoft 365 Copilot, which is excellent if you’re already deep in M365 and willing to do the permissions cleanup it requires
• Pay for ChatGPT Team or Enterprise, which has historically been priced for larger teams
• Use the Claude API directly, which works if you have a developer on staff but isn’t a fit for most non-technical buyers

Claude for Small Business is the offering that fills the middle of that map: native connectors for the tools small businesses actually run — QuickBooks, PayPal, HubSpot, DocuSign, Canva, Google Workspace, and Microsoft 365 — plus a set of pre-built workflows for things like invoice chasing, month-end close, and cash-flow forecasting. Notably, there’s no extra subscription on top of your Claude licenses, and no custom development required.

If you’ve been waiting for a sanctioned, non-developer way to roll Claude out to your team, this is it.

What the KPMG and PwC deals mean for the rest of us

You’re not KPMG. You don’t have 276,000 employees. So why does it matter that they’re deploying Claude across their workforce?

Two reasons.

First, it removes the “is this real” question from the conversation. Until recently, AI in the enterprise was an experimental line item. Now the firms that audit your clients, advise your competitors, and tell their clients what to do with their technology are betting their own operations on this. When PwC tells a mid-market client to look at AI for finance workflows, they’re now speaking from experience, not from a slide deck.

Second, it shifts the talent and tooling market. The big consulting firms will produce playbooks, methodologies, and reference architectures that trickle down. They’ll also hire the people who would otherwise have been available to small businesses for the same work. The window for an SMB to get ahead by adopting deliberately, before the playbooks become commoditized, is right now.

What the Stainless acquisition signals

This one is more technical but it matters for any business that’s planning to use AI for more than chat.

The new wave of AI products are called “agents,” and the easy way to understand the difference between an agent and a chatbot is this. A chatbot answers your question. An agent does the work. You don’t ask the agent how to file an expense report. You hand it the receipt and walk away while it actually files the report, attaches the documentation, and routes it for approval.

Building real agents is hard because they need access to your actual business systems. They have to read your email, write to your file storage, make API calls, and sometimes browse the web on your behalf. Every one of those touchpoints is a place where security, performance, and reliability can break.

That’s why Anthropic paying a reported $300+ million for Stainless matters. Stainless builds the plumbing — the SDKs and connectors that let agents talk to APIs — that hundreds of companies, including Anthropic’s own rivals, relied on. When a vendor buys the toolchain layer of the agent ecosystem outright, it’s telling you where it believes the next several years of value live: not in chat, but in agents wired into real business systems through controlled, audited channels.

If you’ve been waiting for “agents” to graduate from impressive demos to deployable workloads, the infrastructure consolidation happening right now is the strongest signal yet that the industry agrees with you.

What you should actually do this quarter

Translating the news into a real plan looks like this.

Pick a sanctioned stack and commit

The single biggest unforced error we see in small business AI right now is using too many tools without sanctioning any of them. Employees use ChatGPT for some things, Claude for other things, a free trial of Gemini for a third, and a random AI meeting note-taker that nobody approved. The data leaks compound and the productivity gains never materialize because everyone is teaching themselves a different tool.

Pick one. For most M365-centric businesses, that means Microsoft 365 Copilot for in-context productivity and one general-purpose AI for everything else. The general-purpose layer is where Claude for Small Business is now a strong candidate. Pay for it, deploy it, train on it, and ban the rest.

Clean up permissions before you turn on Copilot

This deserves its own emphasis because every Copilot rollout that goes badly goes badly for the same reason. Copilot can only respect the permissions that already exist in your Microsoft 365 tenant. If your SharePoint sites are overshared from years of “share with everyone, we’ll clean it up later,” Copilot will surface that data to anyone who asks the right question. The clean-up happens before the rollout, not after. We covered the full pattern in our AI tools at work post.

Identify two or three concrete workflows, not “everywhere”

The companies that get real value from AI in 2026 are not the ones that deploy it organization-wide on day one. They’re the ones that pick a handful of high-volume, well-defined workflows and automate them deliberately. Customer support email drafting. Sales call summaries. RFP response generation. Marketing copy production. Financial variance commentary. Pick the two or three that map to your highest-cost or highest-volume work, get them right, and then expand from there.

Write down what data is allowed where

Two pages, not twenty. Which AI tools are approved. What data each one can touch. What’s off limits in any AI tool. Who to ask for exceptions. We have a five-bucket framework that works for most SMBs in our AI tools at work post. Use it, modify it, write your own. Just make sure something exists.

Plan for the agent wave before it arrives

The next eighteen months are going to bring a flood of “agentic” tools that promise to take work off your plate by acting autonomously. Some will be transformative. Many will be vaporware. A few will be security disasters waiting to happen. The right posture is curious and patient. Pilot agents in low-stakes workflows. Don’t give them admin credentials. Make sure every agent action has a logged audit trail. Don’t deploy any agent that doesn’t tell you exactly what it did when it’s done.

What the Microsoft side looks like right now

For businesses already deeply invested in Microsoft 365, Copilot remains the path of least resistance for in-context AI. It’s getting better quickly. The latest updates extend it deeper into the Office apps, improve grounding in your tenant content, and expand the workflow templates available for common tasks.

Copilot is excellent at “help me do what I’m already doing, faster.” It’s less suited to “be my general-purpose research and writing partner outside Office.” That’s where having a second sanctioned tool, like Claude for Small Business, makes the stack complete. The two are complementary, not competing, for most SMBs.

What about the cost

Honest answer. For a typical 30-person business that gets serious about this in 2026, you’re looking at $30 to $50 per user per month in AI tooling all-in, plus whatever you spend on training and rollout. That’s real money. The companies that justify it are the ones that pick the workflows where the AI saves measurable hours per week per user. The companies that don’t justify it are the ones that buy the license, send a setup email, and hope productivity goes up on its own.

The productivity gain is real. The default rollout doesn’t produce it. The deliberate rollout does.

The bottom line

AI for small business stopped being theoretical sometime in the last six months. The vendors have built the products, signed the enterprise deals that validate the technology, and lowered the friction for a 30-person company to deploy it the right way. The risk is no longer “we wasted money on hype.” The risk is now “our competitors deployed deliberately while we deployed by accident.”

The next twelve months are about picking your stack, hardening the data side, and putting AI on the two or three workflows where it actually moves your numbers.

If you want help mapping that out for your specific business, book a consultation. We work with companies across the Inland Empire on AI rollouts that actually produce measurable returns, not impressive demos. We’ll look at your workflows, your existing M365 setup, and your data posture and tell you honestly what to do first, what to do later, and what to skip.

The science experiment phase is over. Time to deploy.