When you type a prompt, you summarize. When you speak one, you explain. Wispr Flow captures your full reasoning — constraints, edge cases, examples, tone — and turns it into clean, structured text you paste into ChatGPT, Claude, or any AI tool. The difference shows up immediately. More context in, fewer follow-ups out.

89% of messages sent with zero edits. Used by teams at OpenAI, Vercel, and Clay. Try Wispr Flow free — works on Mac, Windows, and iPhone.

Start flowing free

Five tools worth knowing. One clear first hire. What each is actually for.

Most owners are shopping for the best AI. For once there is a straight answer: for a business operator today, the strongest single hire is Claude. But one hire rarely runs a whole company, and the more useful question is still which tool for which job.

Ethan Mollick framed this better than anyone: a model is not one tool, it is a set of coworkers you can hire. You would not put your most expensive senior associate on the phones. You would not send a brand-new temp to close your biggest deal. Same with these tools. The skill is matching the hire to the work in front of you.

Here is how most operators actually use AI. They open one tool, usually ChatGPT because it came first, and ask it to do everything. Draft the email, read the spreadsheet, research a competitor, write the proposal. It works well enough that they never look further. Well enough is the trap.

Five names cover what an SMB owner needs, and they are not equals. Claude leads. Gemini and ChatGPT are strong generalists with clear lanes. Grok and Perplexity are research specialists. The big three cost about the same at the entry tier: a generous free version, then roughly twenty dollars a month. Price is not the deciding factor. Fit is.

The model that runs your AI-first operating system.MAIN FLOOR

Claude is the first hire, and the gap is not subtle. It writes long-form prose that sounds like a person instead of a press release, it is the strongest of the field at code, and it holds a long document without losing the thread. The work where a sloppy draft costs you money lands on this desk.
What sets it apart is that it runs the whole operation, not just a chat window. Projects hold your business context. Connectors reach into your inbox, calendar, files, and CRM. Cowork does hands-on work across your computer, and Claude Code builds and automates. My entire business, including this newsletter, runs on that stack.
One subscription is really four hires of different seniority. Fable 5, released yesterday, is the new top model for the hardest, longest work, and I am testing it now. Opus 4.8 is the senior partner when being wrong is expensive. Sonnet is the everyday default. Haiku is the fast, cheap one for high-volume steps.

Use Claude for:

Native to everything Google.

Gemini is the one already sitting inside your Google account. If your business runs on Gmail, Docs, Sheets, and Calendar, it reads them where they live with nothing to wire up. NotebookLM turns a pile of documents into briefings, study guides, and podcast-style audio. That native reach is the lane it owns outright.

The announcement to know about is Gemini Spark. Unveiled at Google's developer conference last month, it is a 24/7 assistant that keeps working with your laptop closed: it works across your Google apps and connected tools, checks with you before big actions, and surfaces results when they matter. Rolling out to top-tier subscribers first, and the clearest preview yet of where all of this is going.

Workspace Studio is the quieter release that matters more for operators. It is a no-code builder, included in Workspace business plans, that turns a written procedure into an automation running across Gmail, Drive, Docs, and Chat. Google says it already runs 170 million automated tasks a month. If your operations manual lives in Google, parts of it can now run themselves.

And it has the strongest image and video models of the three. Nano Banana for images. Veo, joined by the new Omni, for video from a plain description. If your marketing needs visuals every week, this bench alone can earn the subscription. It also has the biggest memory of the field: hand it a full contract stack at once.

Use Gemini for:

The one everyone already has.

ChatGPT is the generalist your team has probably already opened. Widest adoption, the most third-party add-ons, the least convincing required. It lives in its own chat, in Custom GPTs (its version of saved instructions), and now directly inside Excel and Google Sheets. Familiarity like that is worth real money on everyday work.

Its image model deserves specific credit. ChatGPT Images is the easiest way to get a usable visual out of one sentence, and it renders text inside images cleanly, which most models still fumble. Product mockups, ad drafts, a quick diagram for a deck, all done inside the same chat your team already uses.

Like Claude, it comes in seniorities. GPT-5.5 is the flagship for hard, long-running work, and its faster Instant version became the default for every user just last month. A Thinking tier sits between them for problems that need real reasoning. Same hiring logic: match the size of the model to the stakes of the job.

The outlier worth knowing is Codex, OpenAI's builder. It writes and ships software from plain instructions, and it has moved fast in recent weeks: a goal mode that keeps working toward an outcome, a preview that builds and hosts small internal tools, control of Windows machines. My own use is narrow. On bigger Claude Code builds, Codex is the second set of eyes that reviews the work.

Use ChatGPT for:

Two more tabs. One shared job: research.

Grok and Perplexity both earn their seat the same way: detailed web research. Hand either one a real question, the kind you would once have spent an hour Googling, and you get a worked answer instead of ten blue links. Neither replaces your main model. They feed it.

The split is access versus receipts. Grok has a live line into X, so it sees what an industry is actually saying right now, before the press writes it up. That is where I send it every week. Perplexity answers with numbered sources you can click and check, which is what you want when the answer goes in front of a client.

Use them for:

If you are only going to run one, run Claude. The exception is real: a shop that lives entirely in Google Workspace and mostly needs help inside it will get further faster with Gemini, and a team already fluent in ChatGPT should not throw that away. But the operator default is the tool that can run the whole operation.

Whichever you run, watch the seniority ditch. Every maker sells sizes, and most operators either overpay, running the top model on simple work, or underdeliver, running the cheap model on the job that needed the careful one. The everyday tier is the right default. Reach up only when being wrong is expensive.

If you run more than one, give each a job. A separate small-business report this month found that small firms increasingly pay for two and three AI services at once. That only pays off if each one has a lane. Otherwise you are stacking subscriptions and calling it a strategy.

And when a vendor pitches you an AI service, ask which model is under it and why. A cheap, fast model doing high-stakes legal or financial drafting is a flag worth raising before you sign. Two issues back I mapped what is actually inside Claude, room by room, in the Claude map. This issue is the zoom-out: the whole street, and which door to knock on for which job.

The takeaway is two lines. If you hire one AI, hire Claude. Then stop shopping for the best and give every tool on your bench the one job it is best at.

Four reads worth your time this week.

Anthropic releases Claude Fable 5, its most capable model yet

Yesterday, Anthropic shipped Claude Fable 5, the first generally available model in its new Mythos class, a tier above the Opus models. The claim: state of the art on nearly every benchmark, with the biggest gains on long, complex tasks. Two operator notes. It is included in the paid Claude plans at no extra cost through June 22, then moves to usage credits, so the next two weeks are the free window to try it. And I am running my own work through it this week. What it actually changes, if anything, lands in a future issue.

Gemini's next generation landed at Google I/O

At its developer conference, Google rolled out a new Gemini generation and Omni, a model that turns text, images, and audio into video. The operator point is not the version number. It is that the model inside Google Workspace got materially better without you installing anything, and the field jumps again every few weeks. You cannot chase every release. You can know which job each tool is for, which is the whole point of today's lead.

Small businesses are now paying for more than one AI

A new JPMorganChase Institute report, summarized this month by the Small Business & Entrepreneurship Council, tracked what small firms actually pay for, from real bank-transaction data. Among small firms that pay for AI services, the share buying just one fell from 89% to 72% between 2019 and 2025, while those paying for three or more went from almost none to nearly one in ten. Read against today's lead: you are probably already running more than one. The question is whether each one has a job, or whether you are just collecting subscriptions.

AI sticker shock hits corporate America

Big companies are getting their AI bills and flinching. A consultant told Axios one client burned through half a billion dollars in a single month after setting no usage limits on its Claude licenses, and other firms are cutting tools they rushed to buy. The cause was rarely the model. It was paying to automate the tasks people disliked instead of the tasks that mattered, with no cap on the meter. The small-business version is cheaper to fix: pick the high-value job first, and watch what each tool costs you to run.

The plug your vendor keeps mentioning, and the questions it should trigger.

You are going to hear "MCP" in every other AI pitch this year. It stands for Model Context Protocol. Strip the jargon and it is a standard plug. It is how an AI model connects to the tools your business already runs: your inbox, your calendar, your CRM, your files.

Think of it like USB. Before USB, every device needed its own special cable. MCP is the AI version of the universal port. When a vendor says "we support MCP," they mean their system can plug into the same tools yours does, without a custom wire built for each one.

That is genuinely useful, and it is also where the risk lives. A plug that reaches your inbox and your customer records is powerful and exposed at the same time.

So when you hear it, ask three things. Which of my tools does this connect to. What is it allowed to do once connected, read only or also write and send. And who maintains the connector when it breaks.

The answers tell you whether MCP is a feature or a liability in your business.

A short thing from how I actually work.

This week's note is an experiment in progress. Anthropic released Fable 5 yesterday (Signal 01 above), and I am not going to summarize a benchmark chart at you.

I am running my own work through it this week instead. The writing, the research, the builds. Same jobs, new hire, and I want to see where it earns the seat and where it does not.

What holds up lands in a future issue. If you run Claude, the free window through June 22 is your chance to form your own opinion before it costs anything.

See you next Wednesday.

Reply and tell me which AI tools you are running right now. It lands in my inbox at [email protected]. I read every reply, and it decides what I write next.

Talk Wednesday.

Marc