The Amazon Let Me Forget About AI. Quito Did Not.

I emerged from the Ecuadorian Amazon last week with a dead phone, four days of unread emails, and a genuine plan to keep not thinking about AI for a while longer. Quito had other ideas.

I was there for Dean’s wedding, with time on either side of it to actually see the city. Shops, markets, restaurants. Somewhere in the middle of all that I started noticing the posters, and I have not stopped thinking about them since.

Dean is still enjoying the days post-wedding, so you’re stuck with me, Tim, this week.

This week:

• Why the same AI poster gets a shrug in Quito and a pile-on in Cincinnati

• A rental quote written from the front seat of a truck

• Salesforce goes all in on Claude, a third of companies stop buying software, and Cisco hands 90,000 people an agent

The Deep Dive

The Amazon Let Me Forget About AI. Quito Did Not.

On my second day in Quito I passed a lunch spot with a poster in the window. A plate of food, lit beautifully, steam curling in a way steam does not actually curl. It was AI. Obviously AI.

Then I kept walking. A billboard for a phone plan. A gym flyer. A tour company. A bakery. A hardware store. I stopped counting.

Here is the part that matters. Nobody cared.

The restaurant was full. Nobody was photographing the bakery flyer to dunk on it. The images were doing the only job a picture in a window has ever had, which is to tell you what is inside.

Run that same poster in Cincinnati and you know what happens. Somebody posts it. Two hundred people who have never eaten there explain that the owner is cutting corners and the food is probably a lie too. It comes down by Friday.

Same image. Same tool. Same five minutes of somebody’s afternoon. Completely different outcome.

Dean wrote in May about the crowd that confidently called a real Monet AI slop. His point was that once a client feels the work came from AI, you lose trust. He was right. What Quito taught me is that it is local. Not a property of the tool. A property of the room.

Before anyone decides that room is less discerning, the numbers say the opposite. KPMG and the University of Melbourne surveyed 47 countries. In emerging economies, three in five people say they trust AI systems, against two in five in advanced economies. Reported benefits, 82 percent against 65. Self reported AI literacy, 64 percent against 46.

Higher trust, higher literacy, higher reported benefit. The reflex to call it slop assumes people using it openly do not know what they are looking at. They know. It does not bother them, which is a choice rather than a deficit.

Now, one thing about my counting. Every poster I flagged that week was one I was able to spot. Anything made carefully, edited, and actually finished went straight past me and got filed as a normal poster. I have no idea how many of those I walked by. Neither do you, about the ones on your own street.

Which points at the real skill, and it is not knowing your audience.

Knowing your market tells you how visible the AI can be. It does not tell you whether the work is any good. The ones I spotted were not obvious because a machine made them. They were obvious because nobody moved off the defaults. Same lighting, same gloss, same composition every model reaches for when you do not ask it for anything else. That look is its own logo now, and what it says is that nobody here spent more than a minute.

So the skill is not avoiding these tools. It is knowing them well enough that the output looks like you. Your colors, your photography, your own storefront instead of a generic one. It takes an afternoon and almost nobody bothers, which is why it still reads as effort.

Two questions, then. Where does your market actually sit, and have you tested that or assumed it? And when you use these tools, are you shipping your own work or the defaults?

Reply if you want help with either. That is a conversation we have most weeks.

From the Field

The Quote That Got Written in a Truck

One of our clients rents deep foundation drilling rigs. Big machines, long contracts, and a sales team that is almost never at a desk.

Quoting one of those machines means knowing which unit is free, where it is, when it comes back, what it rents for, and what terms apply to the job. All of that exists in their system. Getting it out used to mean being somewhere with a real computer, or calling two people who might not pick up. Fifteen to thirty minutes, most of it spent chasing availability.

Now a rep opens Claude on the road and asks for a quote. It pulls the unit and the customer, builds the document, and hands it back. Under two minutes, in their words.

He had a job coming up in hard rock, the kind that beats up a machine. So he went into the skill and added his own terms and conditions to cover it. Nobody asked him to. He was not trained on it as a feature. He noticed the standard terms did not fit the job and changed them.

That is the whole thing. We did not build a quote button. We built something a salesman could adjust when reality did not match the template, and then a salesman adjusted it.

A human still reads every quote before it reaches a customer. That one is still sitting in his drafts while he thinks about the pricing, which is exactly right.

Why this travels: your quoting process probably lives in three places and one person’s memory. The win is not the speed. The win is that the person who actually knows the job gets to shape the output instead of filing a request and waiting.

AI News Worth Your Time

Three Stories From This Week

Salesforce made Claude a first-class citizen

On August 26, Salesforce and Anthropic announced Claudeforce. Two halves to it. Salesforce in Claude is a plugin with 37 prebuilt sales skills covering meeting prep, deal health and pipeline review, which build live dashboards from Salesforce and Slack data and route any action back through Salesforce so your business rules still apply. Claude in Salesforce puts Claude underneath Agentforce as the reasoning model. Separately, Claude becomes the default model in Slack, powering Slackbot and Claude Tag. Pilot customers have it now, open beta is this month, and pricing has not been announced.

My take: this is Salesforce catching up. HubSpot and most of the other CRMs have had a Claude connector for a while, and Salesforce users have been living on stopgaps the whole time, exporting reports, pasting records into a chat window, or paying somebody to build a bridge in the middle. Now the two actually talk to each other. If you are on Salesforce and you gave up on connecting it to Claude a year ago, that was the right call then and it is worth another look this month. READ THE SALESFORCE ANNOUNCEMENT →

A third of companies killed a software purchase because they could just build it

McKinsey published its State of AI global survey on August 25, covering 1,719 respondents across 97 countries. The number that jumped out: 32 percent said their organization decided against buying at least one software product or feature because they could build the functionality in-house using agentic coding tools. Among the companies reporting real EBIT impact from AI, it was closer to half.

My take: the question used to be which vendor. Now the first question is whether to buy at all. You are not going to build your own accounting system, and you should not try. But the fourth subscription that does one small thing for two hundred a month is a genuinely different conversation than it was last year. Worth running the list of what you pay for and asking which line items are really just a form, a rule, and a report. READ THE MCKINSEY SURVEY →

Cisco gave all 90,000 of its employees their own agent

Cisco is rolling out MyAgent to its entire workforce, an agent with persistent memory that works across Outlook, Webex, Jira and SharePoint inside their own governed environment. You give it an objective and it coordinates the steps, with a human still accountable for the outcome.

My take: the headline number is the scale, but the part worth stealing is the word personal. The most common complaint we hear from people using these tools at work is that the thing does not know who they are, so every task starts with re-explaining your role, your territory, your customers. Cisco solved that with a platform. You can solve it in ten minutes by having your team tell Claude who they are once and letting it remember. That gap is smaller than it looks. READ THE CISCO POST →

That’s it for this week. Dean has it back next Tuesday, and I go back to AI Under Management, the newsletter we write for wealth and asset management firms, where the second edition goes out Thursday.

If this was useful, forward it to whoever on your team is in charge of making things look good. That’s who this is for.

And if you want to talk about what AI could do for your business, not in theory, but in your workflows, with your team, on your timeline, reach out at [email protected]. That’s what we’re here for.