Latest / Connor with Honor: AI, Real Estate and Real Life / AI Promises, Accountability, and Real Use

Transcript

AI Promises, Accountability, and Real Use · Connor with Honor: AI, Real Estate and Real Life

• 0:00
  SPEAKER_00: Your AI can look productive while making everybody around it work harder.
• 0:04
  SPEAKER_00: A Reddit commenter said that their manager handed them about five pages of AI generated material to turn into a presentation.
• 0:12
  SPEAKER_00: Their complaint, cleaning it up and checking it took roughly three times as long as building the presentation from their existing documentation.
• 0:20
  SPEAKER_00: That comment showed 620 votes, and when I researched this episode, a reply showing 304 votes described catching errors in a coworker's unreviewed AI output only to hear the manager excuse because somebody had caught the mistakes.
• 0:37
  SPEAKER_00: Those are unverified personal accounts, not a productivity study.
• 0:40
  SPEAKER_00: But listen to what people are angry about.
• 0:42
  SPEAKER_00: Somebody got credit for producing work.
• 0:44
  SPEAKER_00: Somebody else inherited the responsibility for making it useful.
• 0:48
  SPEAKER_00: If you're a business owner, that complaint deserves your attention more than you buy another AI subscription.
• 0:54
  SPEAKER_00: I'm Conor McIver.
• 0:55
  SPEAKER_00: This is AI with Honor.
• 0:56
  SPEAKER_00: I build this technology.
• 0:57
  SPEAKER_00: I want the useful parts to succeed.
• 1:00
  SPEAKER_00: My background as an LAPD officer and instructor also make me unwilling to confuse a confidence statement with a supported conclusion.
• 1:07
  SPEAKER_00: So today we're going to talk about that anger seriously, without letting anger write that verdict.
• 1:13
  SPEAKER_00: We have actual comments from Reddit and Hacker News, including highly upvoted responses.
• 1:19
  SPEAKER_00: They aren't a representative survey, and popularity doesn't prove the claims.
• 1:23
  SPEAKER_00: They just give us questions worth investigating.
• 1:25
  SPEAKER_00: So start with the phrase everybody's throwing around, AI slop.
• 1:29
  SPEAKER_00: Some people use it to dismiss anything made with AI, and for me, that's too broad.
• 1:34
  SPEAKER_00: I care whether the work is accurate, useful, appropriate, and owned by somebody willing to answer for it.
• 1:40
  SPEAKER_00: Human can produce garbage.
• 1:42
  SPEAKER_00: AI can produce excellent work.
• 1:44
  SPEAKER_00: The trouble starts when we celebrate the amount generated and stop examining what the recipient can actually use.
• 1:51
  SPEAKER_00: On Hacker News, in a discussion showing 834 points and 734 comments, a person running a creative community described the extra work of rejecting AI-generated accounts and content.
• 2:04
  SPEAKER_00: I can't verify their operation, but I can recognize the business question their complaint raises: where did cost go?
• 2:11
  SPEAKER_00: A sender might produce more material cheaply while the receiver spends longer checking, filtering, correcting, or deleting it.
• 2:19
  SPEAKER_00: Now put that time on that same worksheet before announcing a breakthrough.
• 2:22
  SPEAKER_00: Picture a hypothetical contractor receiving a beautiful estimate draft.
• 2:27
  SPEAKER_00: The layout looks professional.
• 2:29
  SPEAKER_00: The scope includes a service the company doesn't offer.
• 2:32
  SPEAKER_00: The payment schedule contradicts the owner's instructions.
• 2:35
  SPEAKER_00: The consumer address is wrong.
• 2:37
  SPEAKER_00: Getting a polished document quickly has value only if the whole process, including connection, beats the alternative.
• 2:43
  SPEAKER_00: Now picture the better version.
• 2:45
  SPEAKER_00: Approved service details go in.
• 2:47
  SPEAKER_00: Missing information is flagged.
• 2:49
  SPEAKER_00: The owner reviews the scope and numbers.
• 2:51
  SPEAKER_00: The result is a clear estimate the customer can understand.
• 2:55
  SPEAKER_00: There's a genuine opportunity there.
• 2:58
  SPEAKER_00: It requires less theater and more checking.
• 3:01
  SPEAKER_00: That brings us to a different frustration.
• 3:03
  SPEAKER_00: People selling the experience, the appearance of expertise.
• 3:08
  SPEAKER_00: In a January Reddit discussion whether AI would make software companies worthless, a response showing 204 votes pointed to years of bug fixes.
• 3:17
  SPEAKER_00: Customer understanding and earned trust.
• 3:20
  SPEAKER_00: Generating an application doesn't automatically reproduce all that experience.
• 3:24
  SPEAKER_00: It's an argument about the work surrounding the output.
• 3:27
  SPEAKER_00: It isn't proof that newcomers can't build anything valuable.
• 3:31
  SPEAKER_00: I want newcomers building.
• 3:33
  SPEAKER_00: I want the small business owner who can't afford to explore an idea to have a way to try it.
• 3:39
  SPEAKER_00: I also want them to know when they need help.
• 3:41
  SPEAKER_00: Producing a deliverable and possessing that skill to judge that deliverable are separate accomplishments.
• 3:47
  SPEAKER_00: If you can't explain what the system is supposed to do, recognize when it fails or recover the business when it stops.
• 3:54
  SPEAKER_00: You've got more learning to do.
• 3:55
  SPEAKER_00: That applies whether you wrote the code yourself or asked AI to help.
• 3:59
  SPEAKER_00: Here's a useful test for anybody learning this stuff.
• 4:02
  SPEAKER_00: After AI gives you a result, explain it without looking at the answer.
• 4:06
  SPEAKER_00: What facts did it use?
• 4:08
  SPEAKER_00: What did it assume?
• 4:09
  SPEAKER_00: Where could it be wrong?
• 4:11
  SPEAKER_00: Change an input and predict what should happen, then check.
• 4:15
  SPEAKER_00: You're building judgment instead of collecting outputs.
• 4:18
  SPEAKER_00: For an experienced operator, use that same approach to move faster through familiar work.
• 4:22
  SPEAKER_00: Your knowledge of the business is part of the advantage.
• 4:25
  SPEAKER_00: Keep exercising it.
• 4:27
  SPEAKER_00: Now let's talk money because the comments get heated there too.
• 4:31
  SPEAKER_00: In the same software discussion, the agency participant described clients expecting a project to drop from$2,000 to$200 because AI was involved.
• 4:41
  SPEAKER_00: That's their account, not a market price survey.
• 4:43
  SPEAKER_00: And I understand the customer's question: if the work became easier, what am I paying for?
• 4:49
  SPEAKER_00: I understand the provider's question, too.
• 4:52
  SPEAKER_00: Who's paying for the integration testing support and fixing it when something changes?
• 4:56
  SPEAKER_00: Both questions determine answer without anybody getting insulted.
• 5:00
  SPEAKER_00: A seller shouldn't hide behind technical language to protect a price they can't explain.
• 5:05
  SPEAKER_00: A buyer should not assume the cost of generating a draft represents the cost of delivering dependable service.
• 5:13
  SPEAKER_00: As for the actual work, what will exist when this is finished?
• 5:16
  SPEAKER_00: What systems will it connect?
• 5:18
  SPEAKER_00: What's going to be tested?
• 5:19
  SPEAKER_00: What ongoing work is included?
• 5:21
  SPEAKER_00: What cost extra?
• 5:22
  SPEAKER_00: Who owns the accounts and records?
• 5:24
  SPEAKER_00: What happens if we stop working together at any point?
• 5:28
  SPEAKER_00: A fair proposal should make those answers easier to find than the glowing promise at the top.
• 5:33
  SPEAKER_00: Then investigate the benefit.
• 5:35
  SPEAKER_00: If somebody promises more appointments, ask whether they mean requests, confirmed appointments, attended appointments, or paying customers.
• 5:44
  SPEAKER_00: Those are all very different outcomes.
• 5:46
  SPEAKER_00: If they promise hours saved, ask whose hours and whether checking is included.
• 5:51
  SPEAKER_00: If they promise savings, ask whether that bill actually goes away or whether you simply have more available time.
• 5:59
  SPEAKER_00: Now, available time can be extremely valuable.
• 6:01
  SPEAKER_00: It still needs a purpose.
• 6:03
  SPEAKER_00: Calling an opportunity cash in the bank doesn't put money there.
• 6:07
  SPEAKER_00: The next complaint is over automation.
• 6:09
  SPEAKER_00: A July Reddit comment in an automation discussion described reception style bots failing quietly, getting a slaughter address wrong when the problem discovered only when a consumer or customer complained.
• 6:25
  SPEAKER_00: This was a small comment, not a highly ranked verdict, but the failure mode is worth testing anyway.
• 6:30
  SPEAKER_00: A pleasant voice and a friendly confirmation can make an unfinished task feel completed.
• 6:35
  SPEAKER_00: You don't want that customer discovering the difference while standing outside a locked office.
• 6:40
  SPEAKER_00: Anthropics January 2026 guidance on evaluating agents makes a useful distinction.
• 6:46
  SPEAKER_00: Inspect the actual outcome, not merely what the agent says happened.
• 6:50
  SPEAKER_00: In business terms, look for the appointment in the correct calendar.
• 6:53
  SPEAKER_00: Look for the request in the correct inbox.
• 6:55
  SPEAKER_00: Check whether the response person received it.
• 6:58
  SPEAKER_00: A cheerful sentence saying everything is handed doesn't establish that the underlying action actually succeeded.
• 7:05
  SPEAKER_00: That's where my investigator's lens belongs.
• 7:07
  SPEAKER_00: Follow that claim back to an observable record.
• 7:11
  SPEAKER_00: Let's put all this into realistic, hypothetical business situation.
• 7:15
  SPEAKER_00: A local service owner is out on a job.
• 7:18
  SPEAKER_00: An inquiry arrives after hours.
• 7:21
  SPEAKER_00: Instead of trying to automate the entire business, the owner defines a narrow task.
• 7:25
  SPEAKER_00: Explain approved service information, collect the request, and give the customer an accurate next step.
• 7:31
  SPEAKER_00: The AI identifies itself, it doesn't vent availability, promise an unapproved price, or pretend a request is a confirmed booking.
• 7:40
  SPEAKER_00: A named person owns that follow-up.
• 7:42
  SPEAKER_00: Now test the awkward calls because ordinary customers don't follow your demonstration script.
• 7:49
  SPEAKER_00: The caller changes the address halfway through.
• 7:51
  SPEAKER_00: They ask for a service you don't provide.
• 7:53
  SPEAKER_00: They want a person.
• 7:54
  SPEAKER_00: They become frustrated.
• 7:55
  SPEAKER_00: The calendar connection stops working.
• 7:58
  SPEAKER_00: Ask what the system does in each situation.
• 8:01
  SPEAKER_00: Then watch it do it.
• 8:03
  SPEAKER_00: If the fallback is a callback request, say that plainly.
• 8:06
  SPEAKER_00: If nobody is available immediately, don't suggest otherwise.
• 8:10
  SPEAKER_00: A truthful limit is more useful than a smooth promise your team can't keep.
• 8:15
  SPEAKER_00: Now here's how I would begin evaluating that idea.
• 8:19
  SPEAKER_00: Take 20 representative past inquiries.
• 8:22
  SPEAKER_00: Remove unnecessary personal details and turn them into test cases.
• 8:26
  SPEAKER_00: Include routine requests, missing information, and situations requiring a human.
• 8:31
  SPEAKER_00: Define the acceptable outcome before running the test.
• 8:34
  SPEAKER_00: Check the records afterwards.
• 8:36
  SPEAKER_00: The small set is an initial screen, not proof of reliability.
• 8:41
  SPEAKER_00: If it looks promising, move it to a limited, supervised pilot with a clear way to stop.
• 8:46
  SPEAKER_00: Keep checking as real cases reveal what your original examples missed.
• 8:51
  SPEAKER_00: Use ordinary arithmetic as well.
• 8:54
  SPEAKER_00: Now here are hypothetical numbers, not a customer result.
• 8:58
  SPEAKER_00: A weekly process takes 180 minutes today.
• 9:01
  SPEAKER_00: With AI handling takes 50 minutes.
• 9:04
  SPEAKER_00: Review takes 40.
• 9:06
  SPEAKER_00: And corrections and upkeep take another 30.
• 9:08
  SPEAKER_00: That's 120 minutes total, leaving 60 minutes potentially available.
• 9:13
  SPEAKER_00: Then account for setup and software cost.
• 9:17
  SPEAKER_00: Decide what happens with that hour.
• 9:19
  SPEAKER_00: More useful conversations, faster estimates, getting home earlier.
• 9:23
  SPEAKER_00: Those all can matter, but if the errors or cost erase the benefit, change the design, or stop.
• 9:30
  SPEAKER_00: You don't owe a tool your loyalty.
• 9:33
  SPEAKER_00: Also compare it with a simpler solution.
• 9:35
  SPEAKER_00: Maybe a real problem is an outdated form, unclear responsibilities, or nobody checking the shared inbox.
• 9:42
  SPEAKER_00: Anthropic's published engineering guidance recommends starting with a simplest approach that works in an adding complexity when needed.
• 9:50
  SPEAKER_00: You don't have to purchase an autonomous agent to prove you understand the future.
• 9:54
  SPEAKER_00: Sometimes better instructions and an ordinary reminder solve the immediate problem.
• 9:59
  SPEAKER_00: Sometimes AI genuinely helps interpret messy requests that a rigid form handles badly.
• 10:05
  SPEAKER_00: Now test that difference in your own setting.
• 10:08
  SPEAKER_00: There's an important human part here.
• 10:10
  SPEAKER_00: So when workers worry about becoming unpaid quality control for somebody else's AI output, don't call them resistant.
• 10:19
  SPEAKER_00: Ask what's landed on their desk.
• 10:22
  SPEAKER_00: When customers say they can't reach a person, don't respond with your automation statistics.
• 10:28
  SPEAKER_00: Try their route through that system.
• 10:31
  SPEAKER_00: When an employee worries about their job, don't pretend a cheerful speech about opportunity answers what happens to their paycheck.
• 10:38
  SPEAKER_00: Involve the people who know where the word starts to get difficult.
• 10:42
  SPEAKER_00: And that is also where good businesses can stand out.
• 10:46
  SPEAKER_00: Help that customer reach a useful answer.
• 10:49
  SPEAKER_00: Make your proposal understandable.
• 10:51
  SPEAKER_00: Deliver work that you've checked.
• 10:53
  SPEAKER_00: Teach the team how to operate the system and when to override it.
• 10:56
  SPEAKER_00: Keep the benefits visible and the responsibilities clear.
• 10:59
  SPEAKER_00: None of that requires pretending AI is harmless or useless.
• 11:03
  SPEAKER_00: It requires having enough respect for people to notice when your improvement has made their day worse and enough discipline to fix it.
• 11:10
  SPEAKER_00: So here's the standard I want you to leave with.
• 11:13
  SPEAKER_00: Before accepting an AI promise, identify the person making it, the exact result promise, the evidence behind it, who benefits, how you can check it independently, and apply that to an enthusiastic seller and an angry critic.
• 11:29
  SPEAKER_00: Apply it to me.
• 11:30
  SPEAKER_00: A strong idea should survive a reasonable question.
• 11:34
  SPEAKER_00: A useful system should survive contact with the real work.
• 11:38
  SPEAKER_00: If missed calls or messy follow-up or the problem you actually have, visit hireaivoice.com or honorelevate.com.
• 11:46
  SPEAKER_00: Try the voice demo with fictional details.
• 11:48
  SPEAKER_00: Ask for a person, listen to the next step.
• 11:51
  SPEAKER_00: Then bring me your process and your questions.
• 11:54
  SPEAKER_00: We'll talk about fit and scope before treating a demonstration as a proof of results for your business.
• 12:00
  SPEAKER_00: And then tell me what your own experience has been.
• 12:03
  SPEAKER_00: When did AI or what did AI genuinely improve?
• 12:06
  SPEAKER_00: What did it, when did it create or where did it create more cleanup?
• 12:10
  SPEAKER_00: Leave out customer names and private information.
• 12:13
  SPEAKER_00: Describe the task, the failure or benefit and how you checked.
• 12:17
  SPEAKER_00: That's a conversation worth having.
• 12:19
  SPEAKER_00: The angry comments have a point when they ask who carries the work and who owns the mistake.
• 12:24
  SPEAKER_00: Let's answer with something better than another argument.
• 12:27
  SPEAKER_00: Building something useful.
• 12:28
  SPEAKER_00: Show what happened.
• 12:30
  SPEAKER_00: Keep your word.
• 12:31
  SPEAKER_00: That's AI with honor.
• 12:32
  SPEAKER_00: Adcana with honor.
• 12:34
  SPEAKER_00: Over and out.