AI Promises, Accountability, and Real Use

The Machine

Your AI can look productive while making everybody around it work harder. Somebody gets credit for producing the work. Somebody else inherits the job of making it useful. If you run a business, that complaint deserves your attention before you buy another AI subscription.

I build this technology. I want the useful parts to succeed. My background as an LAPD officer and instructor also makes me unwilling to confuse a confident statement with a supported conclusion. So today we take the anger about AI seriously, without letting the anger write the verdict.

Why are people angry about AI at work?

Because the cleanup lands on somebody who didn't choose the tool. One public Reddit comment described a manager handing over about five pages of AI generated material to turn into a presentation. The commenter said cleaning it up and checking it took roughly three times as long as building the presentation from the existing documentation. A reply described catching errors in a coworker's unreviewed AI output, then hearing the manager use the fact that somebody caught the mistakes as the excuse.

Those are public comments. They're unverified personal accounts, not a productivity study, and how many people upvoted them proves nothing about whether they're true. I use comments like these the way I'd use a tip. They tell you where to look. They don't tell you what you'll find.

Is "AI slop" a fair label?

It's too broad. Some people use it to dismiss anything made with AI. I care whether the work is accurate, useful, appropriate, and owned by somebody willing to answer for it. Humans can produce garbage. AI can produce excellent work. The trouble starts when we celebrate the amount generated and stop examining what the recipient can actually use.

In one public Hacker News discussion, a person running a creative community described the extra work of rejecting AI generated accounts and content. I can't verify their operation. I can recognize the business question it raises. Where did the cost go? A sender can produce more material cheaply while the receiver spends longer checking, filtering, correcting or deleting it. Put that time on the same worksheet before you announce a breakthrough.

What does a polished AI draft actually cost?

The draft plus every minute of correction. Picture a hypothetical contractor receiving a beautiful estimate draft. The layout looks professional. The scope includes a service the company doesn't offer. The payment schedule contradicts the owner's instructions. The customer address is wrong. A polished document delivered fast has value only if the whole process, correction included, beats the alternative.

Now picture the better version. Approved service details go in. Missing information gets flagged. The owner reviews the scope and the numbers. The customer gets a clear estimate they can understand. There's a real opportunity there. It takes less theater and more checking. I made the same point about checklists in a polished AI checklist can still miss the task that matters.

Can AI replace experience?

It can produce a deliverable. It doesn't hand you the judgment to evaluate it. In a public Reddit thread asking whether AI would make software companies worthless, one response pointed to years of bug fixes, customer understanding and earned trust. Generating an application doesn't automatically reproduce all of that. That's an argument about the work surrounding the output. It isn't proof that newcomers can't build anything valuable.

I want newcomers building. I want the small business owner who could never afford to explore an idea to have a way to try it. I also want them to know when they need help. Producing a deliverable and having the skill to judge it are separate accomplishments. If you can't explain what the system is supposed to do, recognize when it fails, or recover the business when it stops, you have more learning to do. That's true whether you wrote the code yourself or asked AI to help. Your own playbook is the difference, which is the whole argument in an AI idea is a starting point.

You're building judgment instead of collecting outputs. Experienced operators can use the same approach to move faster through familiar work. Your knowledge of the business is part of the advantage. Keep exercising it.

Should AI make professional services cheaper?

Both sides have a fair question. In the same public software discussion, an agency participant described clients expecting a project to drop from 2,000 dollars to 200 dollars because AI was involved. That's one person's account, not a market price survey.

I understand the customer's question. If the work got easier, what am I paying for? I understand the provider's question too. Who pays for the integration, the testing, the support, and the fix when something changes? A seller shouldn't hide behind technical language to protect a price they can't explain. A buyer shouldn't assume the cost of generating a draft is the cost of delivering dependable service.

A fair proposal makes those answers easier to find than the glowing promise at the top. Most of the AI lead gen pitches in your feed fail this test on line one. I broke that down in AI lead gen offers, mostly nonsense.

How do you check an AI promise of more appointments or saved hours?

Make them name the exact outcome. If somebody promises more appointments, ask whether they mean requests, confirmed appointments, attended appointments, or paying customers. Those are very different results. If they promise hours saved, ask whose hours, and whether checking is included. If they promise savings, ask whether a bill actually goes away or whether you simply have more available time.

Available time can be extremely valuable. It still needs a purpose. Calling an opportunity cash in the bank doesn't put money there.

What happens when automation fails quietly?

The customer finds out before you do. A public Reddit comment in an automation discussion described reception style bots failing quietly, getting a time slot or an address wrong, with the problem discovered only when a customer complained. It was a small comment, not a highly ranked verdict. The failure mode is worth testing anyway. A pleasant voice and a friendly confirmation can make an unfinished task feel finished. You don't want your customer discovering the difference while standing outside a locked office.

The fix is to inspect the actual outcome, not what the agent says happened. In business terms, look for the appointment in the correct calendar. Look for the request in the correct inbox. Check whether the responsible person received it. A cheerful sentence saying everything is handled doesn't establish that the underlying action succeeded.

What does a realistic first AI project look like?

Narrow. Here's a hypothetical. A local service owner is out on a job. An inquiry arrives after hours. Instead of trying to automate the whole business, the owner defines one task: explain approved service information, collect the request, and give the customer an accurate next step.

The AI identifies itself. It doesn't invent availability, promise an unapproved price, or pretend a request is a confirmed booking. A named person owns the follow up. Then test the awkward calls, because ordinary customers don't follow your demonstration script. The caller changes the address halfway through. They ask for a service you don't provide. They want a person. They get frustrated. The calendar connection stops working. Ask what the system does in each situation, then watch it do it.

If the fallback is a callback request, say that plainly. If nobody is available right now, don't suggest otherwise. A truthful limit is more useful than a smooth promise your team can't keep. I walked through that exact call in stop losing service leads after voicemail.

How should a small business start evaluating an AI tool?

With your own past cases, before any live customer touches it.

The small set is an initial screen, not proof of reliability. Keep checking as real cases show you what your original examples missed.

How do you calculate real AI time savings?

Count every minute, not just the fast part. Here are hypothetical numbers, not a customer result.

That's 120 minutes in total, leaving 60 minutes potentially available. Then account for setup and software costs, and decide what happens with that hour. More useful conversations. Faster estimates. Getting home earlier. Those can all matter. If the errors or the cost erase the benefit, change the design or stop. You don't owe a tool your loyalty.

Do you need an AI agent at all?

Maybe not. Compare it with the simpler solution first. The real problem might be an outdated form, unclear responsibilities, or nobody checking the shared inbox. Start with the simplest approach that works and add complexity only when you need it. You don't have to buy an autonomous agent to prove you understand the future.

Sometimes better instructions and an ordinary reminder solve the immediate problem. Sometimes AI genuinely helps interpret messy requests a rigid form handles badly. Test that difference in your own setting.

What about the people doing the work?

Ask them before you announce anything. When workers worry about becoming unpaid quality control for somebody else's AI output, don't call them resistant. Ask what landed on their desk. When customers say they can't reach a person, don't answer with your automation statistics. Try their route through the system. When an employee worries about their job, don't pretend a cheerful speech about opportunity answers what happens to their paycheck. A person signs that decision, which is the point of the machine didn't sign the layoff.

This is also where good businesses can stand out. Help the customer reach a useful answer. Make your proposal understandable. Deliver work you've checked. Teach the team how to operate the system and when to override it. Keep the benefits visible and the responsibilities clear. None of that requires pretending AI is harmless or useless. It requires enough respect for people to notice when your improvement made their day worse, and enough discipline to fix it.

The standard for any AI promise

Before accepting an AI promise, identify five things.

Apply it to an enthusiastic seller. Apply it to an angry critic. Apply it to me. A strong idea should survive a reasonable question. A useful system should survive contact with the real work.

Then tell me what your own experience has been. What did AI genuinely improve? Where did it create more cleanup? Leave out customer names and private information. Describe the task, the failure or the benefit, and how you checked. The angry comments have a point when they ask who carries the work and who owns the mistake. Let's answer with something better than another argument.

Video chapters

Build something useful. Show what happened. Keep your word. That's AI With Honor. Let's be careful out there. I'm Connor with honor, and I'll see you in the next one.