Billy Jack Jones

The most important part of an AI operating system is not the model. It is the receipt trail. If an agent cannot show what it did, where it sent it, and what changed, it is not operations — it is theater. Tidegate is being built around approvals, outbox truth, and proof-first execution for that reason.

With Fable in town, I’m redoing my Agent OS. I love using Hermes. Here I’m running GPT 5.5 for critical plumbing, Kimi K2.6 for less technical tasks, and Fable on all the rest in parallel

The AI ops work that matters for small businesses is not another chatbot in a side tab. It is the control plane behind it: what needs review, what an agent can do, what actually ran, and what proof came out. Agents can prepare the work. The send button still needs a gate.

Fractional AI ops is not “we install automations.” That is too small. The real job is closer to: Understand how the business actually runs. Find the owner bottlenecks. Map where revenue leaks. Decide what should be read-only. Build the safest useful workflow. Add approval gates where trust matters. Turn the result into an operating rhythm. Capture proof so the work compounds. The automation is one piece. The operating judgment is the part that makes it useful. Small businesses do not need a pile of disconnected AI tricks. They need someone to turn messy daily work into systems the owner can trust.

A report is only useful if it changes behavior. This is why so many dashboards die. They technically contain the truth. But nobody knows what to do next. A useful report should make the next action obvious: Call these people. Review this exception. Ignore this noise. Approve this draft. Fix this broken sync. Ask this staff member for context. If the owner has to read the whole report, interpret it, remember the history, and decide the next step from scratch, the system did not remove much work. The value is not in displaying data. The value is in compressing the messy business into a better next move.

The automation is not done when it works once. That is the demo standard. The operating standard is higher. Done means: - it runs again tomorrow - someone knows when it fails - it has a safety boundary - it does not leak private data - it leaves a receipt - the owner knows what changed - the team knows how to use the output - the proof gets captured for future reuse That last part matters more than people think. If you build something useful and do not capture the proof, you lose leverage twice. The business loses the lesson. The market never sees the evidence.

@Codie_Sanchez SMB

I was going to charge for this, but f*** it. My team just put together a 40+ page report of everything we’re seeing right now in the small biz economy. What’s inside: → Real market data on SBA lending → Business buyer demand trends → Data on the ownership succession gap → An inside look at what small businesses are doing with AI right now We’ve spent months doing hundreds of surveys and interviewing dozens of owners to put this together. Now it’s yours for free. Just: 1. Like this post 2. Comment “SMB” And I’ll send it over. (Make sure you’re following me so I can DM you.)

If leads are coming in and revenue is not moving, check the pipe. Most owners jump straight to marketing. More ads. New offer. Better creative. Another agency. Sometimes that is the answer. But a lot of the time, the leads are already there. They are just leaking through: - slow response - one-and-done follow-up - no owner visibility - no second attempt - no stale-lead recovery - no clean handoff from inquiry to close That is not a demand problem. That is an operations problem. Before spending more to fill the top of the funnel, make sure the pipe can actually hold water.

The AI role I care about most is not assistant. It is chief of staff. Not in the corporate-title sense. In the operational sense: - read across systems - notice what changed - prepare the decision - protect the owner from noise - escalate exceptions - draft the follow-up - keep receipts - ask before touching anything risky That is different from “answer my question.” A good chief of staff does not wait for the owner to know what to ask. They surface what matters before it becomes expensive. That is the shape AI ops should take for small businesses. Less chatbot. More operating cadence.

Owner sanity is an operating metric. I do not mean that in a soft, motivational way. If the owner has to personally remember every exception, follow-up, report, handoff, and weird edge case, the business is fragile. It may still make money. It may still look successful. It may even grow. But it grows by increasing cognitive load on the one person least allowed to drop the ball. That is not leverage. A useful ops system should reduce owner load in measurable ways: - fewer places to check - fewer things to remember - fewer surprise fires - fewer “did anyone follow up?” moments - fewer decisions made from stale information The owner should still lead. They should not have to be the entire nervous system.

The useful AI work is usually not the flashy part. It is the plumbing. Permissions. Backups. Source separation. Credential hygiene. Failure alerts. Human approval. Duplicate cleanup. Fresh data. Clear ownership. Nobody gets excited about that in a demo. But it is what separates a toy from something a real business can depend on. A local business owner does not care if the system sounds futuristic. They care if it helps them stop losing leads, missing handoffs, and carrying every exception in their head. That means the boring details matter. Especially the boring details.

Just like the ol’ SOP binder 🤓

A small business needs two operating rhythms. One protects cash today. One builds growth tomorrow. The first rhythm watches the business: - leads slipping - customers drifting - invoices stuck - staff handoffs missed - exceptions that need the owner The second rhythm turns the work into leverage: - proof receipts - case studies - content ideas - sales angles - outreach wedges - better systems for next time Most owners are forced to choose. If they protect cash, growth gets ignored. If they chase growth, operations get messy. The point of AI ops is not to add another initiative. It is to make both rhythms lighter. Protect the floor. Build the future.

This is what Tidegate is built for. Not chatbot theater. Actual operator leverage. Private customer name blurred. Everything else is real.

That’s where local businesses leak money. Not because the owner doesn’t care. Because the follow-up pile gets buried. Because staff are busy. Because every lead needs a slightly different response.

Most AI demos are fake. This one happened this morning. I asked our agent if it could take the last 100 Facebook leads, remove DND contacts, remove active members, check each person’s last reply, and text each one accordingly.

Buying more software is not the same as building an operating system. A lot of small businesses have the same pattern: CRM for sales. Scheduling app for appointments. Billing platform for payments. Email tool for campaigns. Spreadsheet for “the real numbers.” Group chat for emergencies. Owner brain for everything else. Then someone adds AI on top and calls it transformation. But if the underlying workflows are messy, AI just makes the mess faster. The real work is connecting the operating logic: What matters? Who owns it? What happens next? What needs approval? What should never be automated? Tools are ingredients. The operating system is the recipe.

Automation without approval gates is just new risk. This is where a lot of AI advice gets dangerous for small businesses. “Let the agent handle it” sounds efficient until the thing being handled is: - customer messaging - ad spend - billing - scheduling - refunds - public content - anything tied to trust The better pattern is staged autonomy. Stage 1: read-only monitoring. Stage 2: draft recommendations. Stage 3: human approval. Stage 4: limited action with logs. Stage 5: broader autonomy only after trust is earned. That is slower than a hype demo. It is also how you keep the business safe. The goal is not to replace judgment. The goal is to stop wasting judgment on things the system can prepare, summarize, and queue for approval.