Three stories from this week show how AI is moving from side tool to business infrastructure.

Anthropic launched Claude for Small Business on May 13. Claude is Anthropic’s AI assistant, similar in category to ChatGPT. This new version is aimed at smaller companies that do not have enterprise AI teams or large implementation budgets.

The product connects to tools small businesses already use: QuickBooks, PayPal, HubSpot, Canva, Docusign, Google Workspace, Microsoft 365, and Slack. Anthropic says it includes 15 ready-to-run workflows across finance, operations, sales, marketing, HR, and customer service. Examples include planning payroll, chasing invoices, reviewing contracts, and kicking off marketing projects.

The most important detail is not the tool list. It is the approval model.

Anthropic says Claude does the work, but the business owner approves before anything sends, posts, or pays. That matters because the fear many business owners have is not just “Will AI work?” It is “Will it act without me?”

This is where small business AI gets practical. The best use case is not replacing judgment. It is reducing the pile of repetitive, after-hours work that keeps owners stuck in admin mode.

That is also why the training component matters. Anthropic and PayPal launched an AI fluency course for small businesses, and Anthropic is taking the program on a 10-city workshop tour. The first stop began in Chicago on May 14, with other stops including Tulsa, Dallas, Hamilton Township, Baton Rouge, Birmingham, Salt Lake City, Baltimore, San Jose, and Indianapolis.

My practitioner read: this is a serious move because it does not assume small businesses need a different dream. It assumes they need a different implementation path.

HubSpot Is Measuring Whether AI Search Sends You Customers

HubSpot launched AEO Sensor on May 14. AEO stands for answer engine optimization. In plain English, it means how your brand shows up inside answers from tools like ChatGPT, Gemini, and Perplexity.

That may sound like marketing jargon, but the business issue is simple.

If someone asks ChatGPT, “Who is the best website designer for a wellness studio in the Hudson Valley?” and your business does not show up, that is a discovery problem. It is similar to not ranking on Google, except the customer may never see a list of links at all.

HubSpot’s launch came with a data point worth paying attention to. According to HubSpot’s internal data, ChatGPT sent businesses its lowest volume of referral traffic in April 2026 compared with the previous 12 months. The dashboard also tracks volatility across AI-generated brand mentions, citations, citation types, and referred traffic.

This is the shift business owners need to understand: AI tools are not just another traffic source. They can also be a traffic blocker.

People ask the question. The tool gives the answer. The customer moves on.

That does not mean SEO is dead. It means discovery is splitting. Some customers will still search Google. Some will ask AI. Some will ask friends, communities, or private groups because they trust people more than search results.

The practical move is not panic. It is testing.

Ask ChatGPT, Gemini, and Perplexity the questions your ideal customer would ask. See who gets mentioned. See what sources the tools cite. See whether your business is invisible, misrepresented, or accurately described.

That is your starting point.

Andrej Karpathy Joining Anthropic Is a Signal About the Model Race

On May 19, Reuters reported that Andrej Karpathy joined Anthropic. Karpathy was part of OpenAI’s founding team, later led AI work at Tesla, and has become one of the most recognizable AI educators and researchers in the field.

At Anthropic, he is joining the pretraining team. Pretraining is the large-scale training process that gives a model its core knowledge and capabilities before it is adapted for specific uses. In simpler terms: it is part of the work that shapes what Claude can understand and do.

This story is not directly about small business. It is not a tool you can sign up for today. But it matters because talent moves are product signals.

When someone like Karpathy joins Anthropic, it tells us where some of the field’s most influential researchers think meaningful work is happening. It also reinforces that Anthropic is not just trying to sell Claude as a workplace assistant. It is still competing at the frontier model layer, meaning the underlying systems that all the downstream tools depend on.

For business owners, the takeaway is not “follow every AI researcher.” That would be exhausting.

The takeaway is that the tools you use are being shaped by a very concentrated talent war. The companies that attract the strongest researchers can improve faster, build more capable models, and push those capabilities into the tools your team eventually uses.

This is why AI vendor selection is not only about features today. It is also about where the product is likely to be in 6 to 12 months.

What This Means

The through-line this week is that AI is moving into the operating layer of business.

Anthropic is packaging AI for back-office work. HubSpot is measuring how AI changes customer discovery. Anthropic’s Karpathy hire shows the model race underneath those tools is still accelerating.

That is the layer most business owners need to watch.

Not every new chatbot feature matters. Not every model update deserves your attention. But when AI starts touching payroll, invoices, search visibility, referrals, customer service, and the systems your team already uses, it stops being a side tool.

It becomes infrastructure.

And infrastructure needs a plan.

What Business Owners Should Do

Observe your repetitive work before buying another AI tool.

Look for tasks that are high-volume, repeatable, and low-risk. Invoice reminders, meeting summaries, basic lead intake, and first-pass content drafts are better candidates than emotional customer conversations or final financial decisions.
Test your AI search visibility this week.

Ask ChatGPT, Gemini, and Perplexity the questions your customers would ask. Track whether your business appears, which competitors appear, and what sources get cited.
Do not automate actions without approval.

For now, the safest pattern is “AI drafts, human approves.” Anything that sends, posts, pays, refunds, schedules, or changes customer records should stay under human review.
Watch talent moves, but translate them into product questions.

Karpathy joining Anthropic is not something a business owner needs to obsess over. But it is a reminder to ask: which AI vendors are actually improving, and which ones are just marketing?
Treat AI like infrastructure, not a subscription.

A subscription is something you try and cancel. Infrastructure changes how work moves through your business. That means it needs ownership, rules, training, and review.

Chantal Emmanuel is the co-founder of BAMPT, an AI automation systems company for service businesses, and the CTO of LimeLoop. She writes This Week in AI for business owners who want the practitioner read on what AI is actually doing to the way we work.

Work with Us

Claude for Small Business, AI Search, and the Talent War Behind the Tools