The Deerpath Co. newsletter

Before you let an AI agent into your inbox

Tue, 13 Oct 2026 12:00:00 GMT

You wouldn't hand a new hire the master key on day one. So why give an AI agent the keys to your whole inbox?

AI agents don't just answer questions anymore, they do the actual work. In ChatGPT Business, agents can read and send email from shared Outlook mailboxes and change shared calendar events [1]. Claude can connect to Gmail, Calendar and Drive to search and send emails and manage your calendar [2]. Microsoft's Copilot can take on multi-step work in Outlook and other Microsoft apps [3], and Google announced its own agent for Gmail, Drive and Docs on October 8 [4].

Plenty of owners are already using them. In an SBE Council survey of 516 small employers, 58% said they use AI agents regularly or occasionally, and another 14% are planning to [5].

I think agents are worth using. But the real risk usually isn't the AI fumbling a reply. It's handing it more keys than the job needs.

What too many keys looks like

On October 8, researchers at Zenity Labs showed how one prompt sent to one public-facing AI agent on Amazon's cloud let them grab its credentials and reach other agents in the same account, including private conversations and secrets [6] (a research demo, not a real breach). The lesson learned carries over to any agent: it can only leak what it can reach.

The big tools are built with this in mind. Microsoft's Copilot asks for approval before sending email or changing a calendar [3]. Claude asks before each action in Gmail, Calendar and Drive [2]. ChatGPT's workspace agents let you choose which tools and data an agent can use, and ask before sending email [7].

Start with one job

Starting small is safer. It's also how you learn. You need to wrap your head around how these things work before eating the entire AI elephant at once.

Pick one low-risk job. Drafting replies to quote requests, or summarizing a folder of project docs.
Connect only what that job needs. One app or folder, not your whole inbox and Drive.
Keep approvals on. Let it draft, and read everything before it goes out. If your tool can't ask before it sends, pick a job where it never touches Send.
Keep it away from money and promises at first. No payments, customer commitments or employee records.

Once it's done that one job well for some time, give it the next key. Then the next. Along the way, you verify it is working to your satisfaction. The same way you would any new hire.

Want help picking the first job to hand an agent? Grab 15 minutes and let's talk it through.

Book a 15-minute call

Sources

Stop fixing your AI prompts. Start building your AI's skills.

Fri, 09 Oct 2026 12:00:00 GMT

You have a client review on Thursday. You open ChatGPT, paste in the account notes and ask for a summary. Then an agenda. Then talking points. After the meeting, you paste in your notes and ask for a recap. At every step you explain how you like it done. The output is fine. Next week, you start from zero again.

Most AI advice says the fix is better prompts. I think that's the wrong place to put your effort. Getting good at AI is about building skills, not about getting good at prompting. A prompt is gone the moment you hit enter. A skill stays.

That's not how most small businesses use AI yet. An OECD report on small and mid-sized businesses found generative AI gets used more for simple, one-off tasks than for complex, recurring and important ones [1]. The recurring work is where the time goes.

What do you mean by "skill"?

A skill is a saved set of instructions the AI loads and follows. It holds the steps, the order, your tone, your format, and what the AI must never make up. You set it up once. After that, you just say "run my review prep."

The big tools all have a version of this. Claude calls them Skills, and simple ones can be written in plain text with no code [2]. ChatGPT has Projects, which keep instructions and files together and are on every plan, including Free [3]. Gemini has Gems, which save instructions and reference files and cost nothing to create [4].

You don't have to write it yourself

You don't need to know how to build a skill. Your AI can write it for you, from work you've already done together. What you learn is spotting the repeat and knowing when to ask.

Notice the repeat. Anything you've explained to AI three times is a candidate. Review prep, listing descriptions, renewal check-ins, etc.
Do it once the long way. Work it out in a chat until the output is what you'd actually use.
Ask the AI to write the skill. Something like: "Turn what we just did into a skill I can reuse. Include the steps, my tone, the format, and what you should never make up."
Read it, then save it as a Skill, Project or Gem.
When it misses, tell the AI. Have it update the skill, not just the output.

The first version won't be perfect, and it doesn't need to be. Each fix makes it better, and you never have to make the same fix twice.

If you work in a regulated field, a skill is not a compliance shortcut. FINRA says its existing rules on supervision, communications and recordkeeping still apply when firms use generative AI [5]. Keep a person checking anything a client sees.

Tools will change. Capturing how you work, and improving it every time it misses, is something you learn to do. Once you can do it, it carries over to whatever tool comes next.

If you want to figure out how to start using and building skills tailored to your own business, grab 15 minutes and I'll walk you through it.

Book a 15-minute call

Sources

Talk to your AI. Stop typing to it.

Thu, 08 Oct 2026 12:00:00 GMT

This newsletter started as a voice note. I had the idea while making breakfast, said it out loud to my phone, and it was sitting in my notes before I ever sat down at a keyboard.

Most people still use AI the way they use Google: short, careful questions typed into a box. That's slow, and it makes you leave out the context that gets you a good answer.

Speaking is just faster, and it closes gaps in ideas that may otherwise be lost in translation between your head and your keyboard. In a Stanford study, people entered English text on a phone about 3x faster by voice than by typing, 161 words per minute vs. 53 [1]. To be fair, that was short phrases in a quiet room, not a real workday [1]. Your whole week won't get 3x faster. But the speed difference is real.

It helps most when you're away from the desk. Here's a few things to try this week:

Talk through a client meeting recap on the drive back, then have the AI turn it into a clean checklist.
Ramble a client reply out loud, then ask it to tighten it up.
Brain-dump a business idea while walking the dog.

Tools you can try out, today

You can start for free with the mic button on your phone's keyboard. But nowadays, there are powerful AI-enabled apps that will help clean up your dictations as you make them. Plus they can be used across any app on your computer where there is a place to input text.

Here are a couple that we've been using at Deerpath:

Superwhisper: free with models that run on your own device, so on the free plan nothing you dictate leaves it [2]. Pro is $8.49 a month, adds the AI cleanup, and one license covers Mac, Windows, iPhone and Android [2].
Wispr Flow: works in any app on Mac, Windows, iOS and Android, free up to 2,000 words a week on desktop [3]. Pro is $15 per user a month for unlimited [3]. Transcription always runs in its cloud, so check the privacy settings before you dictate anything sensitive [4].

Right now, we prefer Superwhisper for its privacy feature and lower overall cost.

Speak first, verify second

Do keep in mind that while voice-typing does increase your thought-to-text speed, that's not always a good thing. As always, when AI is involved, it's important to keep a "verify" step in place.

First, make sure you check every name, number, date and price, and that what you've voice-typed is accurate to what you said. These tools are good, but they can still mishear things, and a wrong quote might cost you more than the time you saved. Second, that speed makes it easy to say too much, so be careful what you say out loud. The FTC recommends knowing when an assistant is listening, reviewing how recordings are stored, and deleting old ones [5]. OpenAI says it doesn't train on business workspace data by default [6], but that's no reason to read a customer's card number into it.

Trust but verify. That's a habit you should take with you whenever AI is involved.

If you want help navigating the benefits of using AI-assisted voice typing in your business, grab 15 minutes and I'll show you how I use it day to day.

Book a 15-minute call

Sources

Your business needs an AI operating system

Wed, 07 Oct 2026 12:00:00 GMT

When personal computers showed up, every business eventually needed one. Not because they were trendy, but because that's where the work moved. AI is the same shift, and it's moving just as fast. A Federal Reserve survey post this week says generative AI is being adopted as rapidly as personal computers were in the 1980s [1].

Small businesses are already buying in. In a U.S. Chamber of Commerce survey, 58% of small businesses said they use generative AI in 2025, up from 40% in 2024 and 23% in 2023 [2].

Problem is, most owners use it like a calculator. Open a chat, ask a question, close the tab. Nothing carries over, so every conversation starts from zero. Sure, it's helpful, but it's kind of like using a private jet to commute to the office.

What an AI operating system is

It's one central place where the AI knows your business: your brand, your customers, your numbers, your recurring tasks. You go there to ask questions, get work done, and improve how the business runs. Because it remembers, it gets more useful every week instead of resetting.

The big platforms all started selling some version of this in the last two weeks:

Meta Muse for Small Business connects to tools like Shopify, QuickBooks, Slack and Canva, and is free with usage limits [3]. Meta says it won't post, send messages or buy anything without your approval [4].
OpenAI dots are always-on agents in ChatGPT that keep making progress between conversations [5]. Your first dot is included with Pro or Business Premium [6], which run $100 a month or $100 per user per month [7].
xAI Grok Bot added Team Bots, shared AI teammates that learn as they work, in public beta on Teams and Enterprise plans [8].

Each of these is trying to solve the same problem. They're trying to make their real product (the LLM itself) easier to put to work.

Custom or off the shelf

Off the shelf is the fast start. Custom takes more setup, but it fits how you actually work and you're not locked into one company's pricing or rules. Either one beats having nothing centralized.

Whichever you pick, set the rules before you connect anything. Decide what it can see and what it can do without asking you. Start with one low-risk task you repeat every week, keep approval on for anything that reaches a customer, and give it more access as it earns trust.

If you want help figuring out which setup fits your business, grab 15 minutes and I'll walk you through what I'd build.

Book a 15-minute call