6 min read
The most important AI news for small businesses last week was not a new chatbot. It was plumbing. Three announcements, none of them flashy, made AI assistants safer to hand a login to, easier to teach once and reuse everywhere, and cheaper to run on your own laptop. Plumbing is boring right up until it means you can finally trust the thing you have been nervously testing. Let us walk through all three.
1. 1Password lets AI agents sign in without ever seeing your password
The biggest blocker to letting an AI agent do real admin work (downloading invoices, updating a listing, checking an order) has always been the login. You either paste your password into a chat, which is a terrible idea, or you do that step yourself every time, which defeats the purpose.
1Password’s answer started in July with 1Password for Claude, and it is clever:
- When the agent reaches a login page, 1Password shows you which saved login it wants and why.
- You approve with Touch ID or a similar biometric prompt on your own device.
- 1Password fills in the sign-in form directly. The AI never receives the password or the one-time code.
- The permission covers a single task and expires when that task ends, so the agent has no standing access tomorrow.
- While an agent drives the browser, the 1Password extension switches into a restricted “Agentic Mode” where only the approved logins are reachable.
In late October, 1Password said it is extending this approval flow beyond Claude to more agent products. Industry roundups list Anthropic’s new Claude Duet and OpenAI’s ChatGPT Work among them, but support is arriving integration by integration, so check the specific tool you use before relying on it. Current support covers usernames, passwords, and authenticator codes; passkeys, social logins, and payment cards are on the roadmap.
Why it matters for a solo owner: this turns “the agent can do it, but I have to babysit every login” into “I tap my fingerprint and keep working.” Pair it with the guardrails in our post on rules to set before giving an agent an account and you have a setup you can actually defend to a client.
2. Agent Skills become an open standard under the Linux Foundation
On October 22, Anthropic announced it is donating the Agent Skills standard to the Agentic AI Foundation, a group that operates under the Linux Foundation, the same nonprofit that looks after Linux and many other shared technologies.
What is a skill? A skill is a folder of written instructions, examples, and sometimes small scripts that teaches an AI how to do one specific job your way. Think “how we write a client proposal,” “how we format an invoice reminder,” or “our brand voice for Instagram captions.” The assistant loads the skill only when the task calls for it.
Moving the standard to a neutral foundation is a stewardship change, not a new product. But it is exactly the kind of change that matters over the long run:
- Write once, reuse anywhere. A skill you build for one assistant becomes far more likely to work in others that adopt the standard.
- Less lock-in. If you switch tools next year because of price or quality, your carefully written playbooks can come with you.
- More shared skills to borrow. Open standards attract community libraries, so expect more ready-made skills for common small business tasks.
If this sounds familiar, it is the same story as MCP, the open standard that lets assistants plug into your other apps. Our beginner’s guide to MCP explains that one. MCP connects the AI to your tools; skills teach it how you like the work done.
Why it matters for a solo owner: the time you spend writing down how your business works is now a durable asset rather than a setting buried in one app. Start a “how we do things” document today. It will become your first skill.
3. Flint: a free coding model that runs on your own laptop
The third piece of plumbing is about cost and privacy. On October 21, a startup called Recursive released Flint, a free coding model built to run locally, meaning on your own computer rather than on a company’s servers.
Flint uses a design where only a small slice of the model works on each request (it activates about 2.5 billion of its 40 billion parameters at a time; parameters are the internal settings a model learns during training). In plain terms, that is how a capable model fits on an ordinary laptop without needing a giant graphics card.
You do not need to be a developer for this to be useful. Solo owners write more code than they think:
- A small tweak to a website theme or a contact form.
- A spreadsheet formula or a Google Sheets script that cleans up a client list.
- A quick automation that renames and files downloaded invoices.
With a local coding model, none of that code or the data it touches leaves your machine, and there is no monthly bill or credit meter. The usual caveats apply: local models are generally less capable than the best cloud models, setup takes some patience, and you should test anything it writes on a copy of your file first. If you have never run a model locally, our walkthrough on running a private AI model on your laptop covers the basics.
Why it matters for a solo owner: “free, private, and good enough for small scripts” is a new option that did not exist a year ago.
The pattern behind all three
Put these together and a picture emerges. The AI industry is quietly building the boring infrastructure that makes agents trustworthy for small businesses:
- Identity: agents can act on your accounts without holding your secrets.
- Know-how: your instructions become portable files, not app-specific settings.
- Cost and privacy: capable models increasingly run on hardware you already own.
That is a real shift from a year ago, when every agent demo came with an asterisk about passwords, lock-in, or bills. The asterisks are getting smaller.
Three things to do this week
- If you use 1Password, check whether your AI assistant is on its supported list, and try one low-stakes task, like downloading last month’s invoice from a supplier portal.
- Write one page called “How we do X” for the task you explain to AI most often. That page is your first portable skill.
- Bookmark Flint for the next time you need a small script and would rather not paste your client data into the cloud.
Which of these three would save you the most time? Tell us in the comments, and tell us which login you would hand to an agent first.
