--- title: "How we are getting clients ready for personal agents" description: "Nine parts on what personal agents change. This one is about what we are doing: the scanner we built, our own site going from 39 to 97, what 98 fitness and wellness sites showed us, the order we fix things in, a door for clients' own agents, and what is not live yet." section: AI and Agents author: David Steel published: 2026-10-24T10:42:22.881Z url: https://www.sneeze.it/blog/how-we-are-getting-clients-ready-for-personal-agents tags: ["personal-agents", "agent-ready", "webmcp", "inside-sneeze-it"] --- # How we are getting clients ready for personal agents _Nine parts on what personal agents change. This one is about what we are doing: the scanner we built, our own site going from 39 to 97, what 98 fitness and wellness sites showed us, the order we fix things in, a door for clients' own agents, and what is not live yet._ For nine parts we have described what happens when your customer gets a full-time assistant. It remembers their brief, reads their inbox, shops before they ask, calls your front desk and leaves a thin trail in your analytics. The fair question at the end is what we are doing about it. We have done four things. We built a tool to measure the problem. We pointed it at ourselves first. We pointed it at our industry, which told us the order to fix things in. And we built a door for our clients' own agents. Here is each one, and a plain list of what is live and what is not. ## We started with our own website We build Agent Ready, a free scan at [agentready.sneeze.it](https://agentready.sneeze.it) that scores whether an AI agent can reach a website, understand it, act on it and finish a booking. On October 9 we ran it on the preview of our own new website. It scored 39 out of 100. Grade F. The Act score, the part that decides whether an agent can do anything, was 6. The causes were ordinary. Our robots.txt turned away the assistants a real customer sends, like ChatGPT-User and Claude-User. Our phone and email were drawn by a script after the page loaded, so an agent saw nothing to call. There was no structured data, no sitemap, no llms.txt, and no form an agent could fill in. About an hour of plumbing took us to 89. Then we registered real tools on the page with WebMCP, a proposed browser standard that lets a site hand an agent a short menu of actions. That took us to 97. :::chart column "Agent Ready score for our own website" unit="" Before | 39 Plumbing fixed | 89 Real tools | 97 source: [Agent Ready rescans after each change](/blog/we-scored-our-own-website-39), October 9, 2026 ::: The tools let an agent list our products, get our contact details, check real open times on the calendar, book a walkthrough, send a proposal request, search this blog and run an Agent Ready scan. Each one returns real data or says it could not. If the calendar cannot take a booking directly, the tool says so and hands back a booking link with the details filled in. :::pullquote We would rather an agent hear "I cannot do that" from our site than act on a guess. ::: ## Then we measured the industry A score for one site is an anecdote. So on October 10 we scanned the corporate websites of 100 US multi-location fitness and wellness brands: gyms, boutique studios, med spas, salons, massage and stretch chains, recovery and IV chains, and wellness clinics. Ninety-eight scanned cleanly. We left out our own clients. :::stats 46 | Median score, out of 100 0 | Sites that scored 80 or more 18 of 98 | Sites behind a bot wall an agent cannot pass 14 | Median Act score source: [Sneeze It Agent Ready scan of 98 US fitness and wellness brand websites](/blog/we-scanned-98-fitness-and-wellness-websites), October 10, 2026 ::: Two findings shaped everything since. First, nobody was blocking AI assistants on purpose: of the 82 sites with a readable robots.txt, none blocked the live assistants. The walls were security settings, the "checking your browser" page a person clicks through and an agent cannot. Second, agents could get in but could not do anything. Not one of the 98 lets an agent see class times or book an appointment, and on the 57 sites with a booking or sign-up path the scanner could reach, 30 had a captcha in the way. Budget did not separate the best sites from the rest. Nine of the top ten publish an llms.txt; four of the bottom forty do. That takes about an hour to write. ## The order we fix things in Put our own site and the 98 together and a sequence falls out. It follows the order an agent hits your site in, with the cheap fixes first. :::steps 1. Open the door Check whether a bot wall or robots.txt turns away the assistants people send. Let verified assistants reach your schedule and booking pages. Until this is fixed, nothing else is seen. 2. State the facts in plain text Prices, hours, locations, what is included and cancellation terms, on the page and in structured data, plus an llms.txt briefing. This is the standing brief from [Part 2](/blog/the-agent-already-knows-what-they-want) and the spec sheet from [Part 4](/blog/your-website-has-a-second-reader). 3. Make contact usable by software Phone and email as real links, and every form field labeled. This is the front desk from [Part 8](/blog/when-an-agent-calls-the-front-desk), online. 4. Clear the last step Keep captchas on logins and payments. Take them off free-class and consultation forms, or give agents another way through. 5. Offer real tools, only real ones WebMCP actions such as checking open times or requesting a booking, each tied to a live system. Never a tool that invents availability. 6. Press the booking platform Ask your booking provider when AI assistants will be able to see availability and book. For most businesses, that step is theirs to build. ::: :::flow "How one fix cycle runs" Scan | Agent Ready scores the site and lists fixes in order of points Fix | The web team makes the top changes, usually without a redesign Rescan | The same checks run again, with the evidence shown Repeat | Move to the next step only when this one holds ::: :::figure wide ![A tidy workbench in the back room of a studio where a row of small model storefront doors stands in a line; a craftsperson's hands, seen from behind, fit a new hinge on one door while the finished door beside it stands open with magenta light](/blog/media/709d80d2f2536f49af4896ca.jpg) One door at a time, in the order an agent walks through them. ::: ## A door for the client's own agent The last piece runs the other way. Our clients will have personal agents too, and an owner's assistant will want to ask about that owner's own marketing. So this week we readied our AI connector, the MCP at [mcp.sneeze.it](https://mcp.sneeze.it), for a client's own agent to connect. We wrote about its walls in [Why our AI connector will never show you another franchisee's numbers](/blog/why-our-mcp-never-shows-another-franchisees-numbers). The rules for a client's agent are strict. It sees only that client's own locations. It never sees spend, budget or cost figures. If an ad account is shared with another location, the whole account is refused rather than split. And anything that would take an action or cost money is held until a Sneeze It admin approves it, and expires after 7 days if nobody does. Status, plainly: built and live, no client connected yet. ## What is live, and what is not This series has tried to separate what is measured from what is forecast, so here is our own status in the same terms. :::compare "Live today" "Not done yet" The free Agent Ready scan, no account needed. A second pass in a real Chrome browser that lists the WebMCP tools a site actually registers, running since October 9. Bot wall detection for challenge pages served with an OK status, fixed October 10. Reports written by fixed rules, not by an AI model. Our own site at 97, with real WebMCP tools. Our AI connector, readied for a client's own agent. --- Our own booking still runs through a third-party scheduler, which is the one point we left on the table on purpose. Agent Ready reads the homepage the way an arriving agent would; it does not log in, and it cannot see a booking app that loads inside a separate platform. None of the 98 sites we scanned lets an agent book yet. And no client's agent is connected to our MCP yet. ::: The agents are early too. As [Part 1](/blog/your-customer-just-hired-an-assistant) laid out, these products are new and many are limited. Claude Duet is a research preview for US Pro and Max subscribers through a waitlist, according to [VentureBeat](https://venturebeat.com/technology/anthropics-answer-to-muse-is-here-claude-duet-a-mobile-and-desktop-ai-agent-that-takes-actions-proactively). OpenAI's dots began rolling out to ChatGPT Pro and Business Premium subscribers, per [SiliconANGLE](https://siliconangle.com/2026/09/29/openai-launches-dots-always-on-ai-agents-in-chatgpt-with-their-own-cloud-computers/). Most of your customers do not have one yet. We think that is a reason to start now. The fixes are cheap, they help human visitors too, and they take weeks to get through a web team. :::takeaways "If you read all ten parts" - Personal agents carry a brief, read the inbox, shop ahead and call the desk. Plain, current, specific facts win at every one of those steps. - In our scan of 98 fitness and wellness sites, the median was 46 and no site reached 80. You are probably level with your competitors, not behind them. - Fix in order: open the door, state the facts, make contact usable, clear the last step, then add real tools. - Measure what you can, label what you cannot, and judge marketing on the total. - Run the scan before you spend on anything new. ::: This is the last part of The Personal Agent Shift. To start from the beginning, go back to [Part 1: Your customer just hired an assistant](/blog/your-customer-just-hired-an-assistant). For how the scanner decides each score, read [How Agent Ready checks a website](/blog/how-agent-ready-checks-a-website). :::cta button="Scan your website free" href="https://agentready.sneeze.it" Where does your website stand with personal agents? Agent Ready gives you the same score and fix list we used on our own site and on 98 others, in about a minute. ::: --- #### Sources - Sneeze It, "We ran our own website through Agent Ready. It scored 39," October 2026. [/blog/we-scored-our-own-website-39](/blog/we-scored-our-own-website-39) - Sneeze It, "We scanned 98 fitness and wellness websites. None was ready for agents," Agent Ready scan of October 10, 2026. [/blog/we-scanned-98-fitness-and-wellness-websites](/blog/we-scanned-98-fitness-and-wellness-websites) - Sneeze It, "How Agent Ready checks a website," October 2026. [/blog/how-agent-ready-checks-a-website](/blog/how-agent-ready-checks-a-website) - Sneeze It, "Why our AI connector will never show you another franchisee's numbers." [/blog/why-our-mcp-never-shows-another-franchisees-numbers](/blog/why-our-mcp-never-shows-another-franchisees-numbers) - Agent Ready methodology. [agentready.sneeze.it/methodology](https://agentready.sneeze.it/methodology) - VentureBeat, "Anthropic's answer to Muse is here: Claude Duet," October 21, 2026. [venturebeat.com](https://venturebeat.com/technology/anthropics-answer-to-muse-is-here-claude-duet-a-mobile-and-desktop-ai-agent-that-takes-actions-proactively) - SiliconANGLE, "OpenAI launches dots, always-on AI agents in ChatGPT with their own cloud computers," September 29, 2026. [siliconangle.com](https://siliconangle.com/2026/09/29/openai-launches-dots-always-on-ai-agents-in-chatgpt-with-their-own-cloud-computers/)