How I built a team of AI agents that share one brain

Kev is a team of AI agents built around one business brain. The agents are Grok Bots that live in your business's own Grok account (which needs an X subscription). Kev brings your scattered data into one place, gives it structure, and runs a learning loop, so the business gets a little better every week.

Everything below is live in Kev today unless it's marked Not yet. No customer results yet; those come later.

Built on trust first

I built Kev through a trust lens. AI fails small businesses on trust, not capability. The tools can already write the reply, but an owner cannot hand over a reply with their name on it to something they have not learned to trust.

Two sources shaped how I think about that. Stephen M. R. Covey's *The Speed of Trust* argues that when trust goes up, work gets faster and cheaper, and that trust is earned through behaviour you can see. The NeuroLeadership Institute, founded by David Rock, studies how certainty, autonomy and fairness shape the way we accept change.

So Kev is supervised on purpose. Agents propose, you decide, and trust is earned area by area, the way it is with a new hire. And the brain is one place, because one brain beats a stack of tools.

The real problem: your business data lives in too many places

Customer conversations sit in WhatsApp and email. Contacts sit in a CRM or a spreadsheet. Prices sit in a PDF, invoices in the accounting app, bookings in a calendar, brand files in a shared folder. The processes sit in the owner's head.

These are data silos, and any tool you point at one only knows that corner. The website has never seen the questions customers ask. The social posts know nothing about what sold. So everything waits for the one person who has the full picture: you.

Adding another tool does not fix that. It adds one more place for data to sit.

Why a learning brain is possible for a small business now

What I am describing is business intelligence: one single source of truth that gets sharper over time. Big companies have had it for years. Small businesses mostly have not: it took integration work, upkeep and analysts, and most of what they know was never in a table.

AI changes that. It can pull the facts out of messy material, and AI agents can do the joining up. With an app like Kev, AI integration becomes a connection you switch on. That is the bet I built Kev on.

What fixes it: one source, structure, then a loop

AI for small business only pays off when it works from the whole picture. The fix has three parts, and the order matters.

Bring the data together. Conversations, records, documents and the apps you already use feed one memory of the business.

Give it structure. The business is laid out by area: who you are, what you sell, your customers, your brand, your processes, your decisions. Each job has one owner, decisions are logged, and all work goes through one approval path.

Run a learning loop. Every conversation, correction, result and failed check goes back into that memory, and everything that works from it uses it next time.

Small gains, compounded

The idea behind the loop is not new. British Cycling made the aggregation of marginal gains famous: improve many small things a little, and the improvements add up. Matthew Syed's *Black Box Thinking* makes the related point that teams improve fastest when every failure is captured and learned from, the way aviation treats a black box.

In a business, each edit, unanswered question and failed check becomes a small lesson. Week after week, they add up.

Three pieces, one system

The agents do the work. There is an AI agent for every job: answering customers, sales and the CRM, marketing, images and video, the website, events, the shop, finance, legal, hiring, IT, finding new customers, tenders and your own inbox and diary. They are Grok Bots in your own Grok account, and they hand work to each other. Their job is the reasoning. The models behind those bots sit with Grok, and we do not choose or swap them.

The brain remembers the business. One memory, inside Kev, that every agent reads from and writes back to. What Sales learns, Marketing uses. Research and answers start from that brain: your niche, your requirements and the environment you work in, not from a blank chat.

The Kev app keeps you in charge. In the app you see what needs you, approve or edit what the agents propose, and see the business at a glance. The app's own work (skills, tools, routing, checks) runs on leading AI models through a gateway, so those models can be updated or swapped when a better one fits.

Every agent runs tested skills

A skill is the tested procedure for a job, such as drafting a customer reply or auditing a web page. Each agent runs the skills for its own area and only sees the parts of the brain that area needs.

You message Kev in Grok, and it passes the work to the right agent. Settings shows your agents and the tools each one uses, in plain words.

Kev grows by giving agents new skills. Two recent ones:

Weekly plans. You set the focus for the quarter. Every Monday, a check-in reviews last week and proposes up to five moves for the agents. Your yes hands them off. In a quiet week, Kev says so instead of inventing work.

Event covers. After your approval, EventsKev sets your Luma event's cover image from the Media in your Library. If you need a new one, DesignKev can make it.

How Kev uses Jev

Jev is a decision model from TypeSafe AI. It does not write text. It answers a defined question with a choice, a score or a yes or no, with a probability for each answer. That makes it fast and cheap for the many small judgments a team of Grok Bots makes all day.

So quick sorting and checking happen first, with Jev, and the larger model only runs where real writing or reasoning is needed. If Jev cannot run, each step falls back to what Kev did before. Each new Jev step is tested on labelled examples before it goes live. In my own Kev business, about 5,000 checks over 30 days cost about 31 cents.

The agents spend their effort on the work itself, and you only see what needs you.

Meet the team

Each agent owns one area and hands work to the others. The list below is read live from the Kev app, so it is always current.

The full list of bots and their skills is not available right now. Every bot is on the Bots page.

Kev leads the team. Some jobs stay with people by design: PeopleKev ranks applications but never rejects anyone, and TendersKev prepares bids that a person submits. LegalKev builds your filing and licence obligations for each country from your registration documents, with official sources. It reads contracts for parties, dates, renewals and unusual clauses, and watches for law changes in your countries. It is not legal advice, and a person decides and submits.

From idea to live post: one production line for social media

Separate tools for images, video, captions and scheduling mean every hand-off is a chance for the brand to drift. Kev replaces that stack with one production line, and it starts above the post.

First, a quarter schedule. StrategyKev proposes a content plan: which posts, on which platforms, on which days, up to 300 items. It can mix website articles (including audience spines when you have an SEO strategy), event promotion and social posts, or be social-only. Approving the plan only approves the schedule.

Then each post is built and approved on its own. Keyword research and a viral finder feed each draft: what to say, and which hooks work in your niches, remade as your own creative. You say yes to each post before it is scheduled or goes live.

1. StudioKev makes the creative

StudioKev makes images, carousels, infographics, quote cards, story frames, thumbnails and LinkedIn document posts from your Library's design system. Every image is checked before you see it.

That check learns too. Between 17 and 19 October, five image jobs scored 7 to 9 and passed, yet the check's own notes named mismatches with the brief, and I rejected all five. The new rule: a note that names a mismatch is a failure, whatever the score. A black box moment in practice.

2. Short videos start as a plan you approve

StudioKev first sends a video plan: script, shots, voice, music, cost and preview stills. Only after your yes does it make the clips, voiceover and music, add captions and your brand outro, and check the finished video.

3. SocialKev writes the words

SocialKev writes single posts, a batch for the week or long-form posts. It also writes email newsletters and sequences sent through Beehiiv, and checks your cadence so your channels do not go quiet.

4. Tweak it in Canva if you want

Change a design by hand in Canva and bring it back into Kev for approval.

5. The Scheduler lines it up

Planned posts and newsletters sit in the Scheduler. Images and videos live in Media.

6. You approve, Kev publishes

Each post opens as its own approval page, readable on your phone. What you approve is exactly what goes out, to Facebook, Instagram, LinkedIn, TikTok, YouTube, X and Google Business Profile. Scheduled posts can be cancelled and live posts taken down.

7. Results feed the next post

Post results come back from every connected platform, alongside your email results, and go into the brain. The next batch starts from what the last one taught.

SEO that goes deeper than blog posts

Plenty of AI tools for small business will write you a blog post. That is the thinnest slice of SEO. WebKev runs the whole job, aiming for a website that Google and AI assistants find and quote.

Find what to fix. Audits, keyword research, near-miss search opportunities and tracking checks.

Write what people search for. Articles built around each audience's main page, AI-answer content and local SEO.

Strengthen and check. Fixes go out in batches, checked for links, accessibility, conversion and AI visibility. After publishing, each change is checked on the live site and put back automatically if the check fails. WebKev publishes to WordPress sites and to sites built on Kev's publishing contract.

How the brain keeps learning

The brain is not a static knowledge base. It is self-learning AI in the plain sense, versioned and growing from what happens in the business: customer conversations, questions Kev could not answer (asked of you once, then known), your edits, post results, monthly numbers and failed checks turned into rules. It checks itself for contradictions and gaps, and logs your decisions, so every agent uses a decision once it is made.

You can read, edit or decline every file, compare versions and restore an earlier one.

The brain also does not guess. If nothing in it answers a question, the agent is told so plainly instead of making something up.

How Kev learns from problems and feature requests

When one of your bots runs into something Kev cannot do properly, the Kev team hears about it, without you writing a support ticket.

What is reported on its own. When Kev refuses or fails a bot's request, or a bot reports a failed job, Kev notes which bot and what went wrong. Names, phone numbers, email addresses, amounts, message text and anything your customers wrote are stripped out before anything is stored. Bots can also report problems they see and Kev cannot. The same problem across many businesses becomes one report, with a count, so the Kev team sees what matters most first.

Asking for a feature. If you want something Kev cannot do, your bot first checks whether Kev can already do it. If not, it asks you once: "Should we ask the Kev team to make this a feature for you?" Yes sends your request in your own words. No is remembered, so you are not asked again.

Where you see it. Settings, Sent to Kev lists everything sent for your business and where it stands, from received to fixed or built. When something is fixed or built, you get a note on Home. Another business never sees what was sent for yours.

Problems bots meet are the fastest way to find what to improve, and owners' requests show what to build next.

The app, page by page

Home shows what needs you and this month's results. Inbox holds customer conversations on WhatsApp, Telegram, email and X direct messages, each with a draft reply to send, edit or hold. CRM is a CRM for small business with one record per person, filled in from conversations, with each fact's source and the next step. Brain, Scheduler, Library and Reports hold what Kev knows, what is planned, your brand and every report. Connections links your apps.

What Kev does not do yet

• Ads on Google or Meta (Not yet)
• Instagram and Facebook DMs (Not yet)
• Website chat (Not yet)

You stay in charge

Every customer reply, post, email, website change and video plan is a proposal first. Once you trust an area, you can give it more freedom: small changes go live on their own with an undo, and one undo steps that area back down.

Start with a guided onboarding

Starter at $47, Growth at $197 and Pro at $497 a month differ only in how much use is included (pricing).

The fastest way in is a live guided onboarding: about two hours, starting from $250, with your first month included. You connect your agents, load your business into the brain, connect your social accounts, set up your brand and design system, and make your first posts. You leave with a working system, or we finish it with you. Groups take at most four businesses.

Private one-to-one onboarding adds my AI systems analysis: I map every system and piece of data you run on into Kev and stay with you until every agent you need is running. Priced by team size: up to 10 staff $1,000, 11 to 50 staff $2,750, and 51 to 200 staff $6,750.

Kev, a business brain that keeps learning, with a bot for every job.

**Book your guided onboarding**

Not ready to pay yet? Join a free webinar on the same page and see the brain and the app live first.