Since 1994

What's always true

about these tools.

AI + Agents in Small Business · Myles Bai

AI has flaws

that never go away.


No upgrade fixes them.
Someone has to catch them every time.

At your scale, that someone is you.

The question for every use case:
Can I catch these, every time, at my volume?

REX BLACK · STRIVE FORUM
01 / 06

Confidently wrong.

It's built to produce an answer, not to say I don't know.
When it doesn't know, it guesses in the same voice it uses when it does.

Big teams

Double checks. Citations. Confidence thresholds.

You

Never trust an output you can't verify in thirty seconds. Name. Number. Date. Claim.

REX BLACK · STRIVE FORUM
02 / 06

Drift.

The company that makes the model updates it. Your customers shift. A regulation changes.
Nothing alerts you. It was right last month. It's slightly wrong this month.

Big teams

Monitoring. Scheduled reviews. Eval suites.

You

One recurring calendar appointment. Thirty minutes, monthly, to review a sample of what the tool has been producing.

REX BLACK · STRIVE FORUM
03 / 06

No memory. No paper trail.

Ask it tomorrow why it gave you a specific answer yesterday. It can't tell you.

Big teams

Logs. Audit trails. Prompt and output capture.

You

Your own log. One Google Sheet. Prompt, output, what you did with it. If it touched a customer or a dollar, write it down.

REX BLACK · STRIVE FORUM
04 / 06

Cost scales.


A personal account is $20 a month.
At real business volume the bill can be $400, $4,000, or $40,000 a month.


Big teams

Caching. Batching. Cheaper models for cheaper steps.

You

Set a hard spend cap in the tool's billing settings on day one. Know the volume that trips it before you turn it on.

REX BLACK · STRIVE FORUM
05 / 06

It doesn't know your business.

It knows the internet as of some date. It does not know your customers, your pricing, your SOPs, your voice.

Big teams

Retrieval pipelines. Connected knowledge bases.

You

Give it what it needs every single time. Paste your pricing. Your SOP. The customer's history. Never assume it remembered from last time. It didn't.

REX BLACK · STRIVE FORUM
06 / 06

Privacy.

What you paste is data out the door.
Free tools may store it and train on it. Paid business tools usually don't. You have to check.

Big teams

Private endpoints. DPAs. PII detection.

You

One rule. If you wouldn't email it to a stranger, don't paste it into a free tool.

REX BLACK · STRIVE FORUM
Answering the question

Green. Yellow. Red.


Green. You can catch all six at your volume. Go.
Yellow. Gaps in one or two. Add a habit first, then go.
Red. Too many gaps. Wait.

Red today doesn't mean red forever. It means you need engineering you don't have yet.

REX BLACK · STRIVE FORUM
Where you actually win

You, multiplied.

  • Drafting emails, proposals, social posts
  • Summarizing meetings, calls, documents
  • Writing the SOPs you've been meaning to write
  • Brainstorming offers, names, positioning
  • Research on companies, competitors, regulations
  • A thinking partner when you're alone with the problem

All six flaws are catchable here. You see every output before it leaves your hands.

REX BLACK · STRIVE FORUM
The Phase 2 decision

Phones. Chats. Intake. Systems.


Real value. Higher burden to watch. Higher cost when it's wrong.


Earn the right to automate this. First prove the workflow works with a human in the loop.

REX BLACK · STRIVE FORUM
The Monday drill

One task. Two weeks. Tally sheet.


Pick one repetitive task eating 5+ hours a week.

  1. AI drafts. You review. You send.
  2. Keep a tally. How often is the draft good with no edits?
  3. Keep a note. Anything it got confidently wrong.

At two weeks you'll know two things. Whether this task is a real automation candidate. And which of the six flaws actually show up in your work.

REX BLACK · STRIVE FORUM
Close

These tools are real.

The flaws are real.


No upgrade fixes them.
Someone has to catch them.
At your scale, that someone is you.

REX BLACK · STRIVE FORUM
Since 1994

Thank you.

Myles Bai · Rex Black Inc. · rexblack.com

Open cold. Thirty seconds. "James just showed you what's possible on a Saturday. I'm going to show you the six flaws that are always true about these tools, and how to decide what to turn on Monday morning." Land. Breathe. Move on.

Read line by line. Beat after "never go away" and again after "catch them." The honest message: no new model and no "production" system makes these disappear. Big companies build layers of engineers to watch for them. You don't have that layer. Your habits are the layer. The word "catch" is intentional. It frames this as standing watch, not as solving. Then pose the question. Don't answer it yet. Tell them to hold it in their head through the next six minutes.

This is the one everyone's been burned by. Get a show of hands: "Who has had AI make up a fact with a straight face?" Every hand goes up. Normalize it. It's not a bug, it's the shape of the tool.

This one is invisible, and that's what makes it dangerous. A prompt that worked in January may quietly stop working in April because a new model shipped, not because anything on your end changed. The habit here is the fix.

This matters the first time something goes wrong. A customer calls upset. You need to reconstruct what happened. Without a log you have nothing. Keep the log. Five seconds per row.

Real story. A founder plugs AI into their phones, gets excited about it routing four hundred calls overnight. Bill arrives: twenty-eight hundred dollars that week. Every vendor has spend caps. Set them on day one.

The assumption that it "learned" your business from a conversation last week is one of the biggest causes of confidently-wrong output. Every conversation is fresh. Feed it what matters, every time.

Customer data. Contracts. Financials. Health info. Use a business-agreement tool or don't use AI for that step. Simple rule, works forever.

This is the moment the question from slide 2 gets answered. This is not "small business can't do AI." It's "know what you're signing up for when you turn it on." Red today doesn't mean red forever. It means the engineering burden is higher than what you can absorb alone right now.

This is where the room gets hopeful again. Energy up. The highest-leverage use of AI for a solo founder is not a customer-facing robot. It's removing the fifteen hours a week you spend writing, summarizing, and researching. The review step is automatic. It's you, reading the draft before sending. That single habit catches five of the six flaws.

Not forbidden. Just later. The hand-off to a customer is where "confidently wrong" stops being embarrassing and starts being expensive. Treat customer-facing automation as a Phase 2 decision, not a Saturday project.

This is the actionable close. Every person in the room leaves with one thing to do. No tools to buy. No consultant to hire. Two weeks, one task, a tally sheet. This is how they learn the six flaws in their own business, for free, without betting anything.

Hold the last line. "Know the six flaws that never go away, and you'll know what you can safely use these for today, and what to wait on until your business can support the engineering to watch for them. That's the whole game." Then stop. Hand back to James for Q and A.

Contact slide. Keep it up while James runs Q and A.