The End of the Yes Reflex

The End of the Yes Reflex. For years, IT's real job was to say yes to whatever the business asked for, and that worked because a bad tool only hurt the team that bought it. AI changes that. An agent or a connector reaches across every system, so the damage from a bad yes now touches the whole company. This week: why AI demand needs one review process instead of a reflex, the three things that break without one, the government switching an AI model off and back on, and where the market is actually putting its money.

Read the full issue: https://quanta-bits-newsletter.beehiiv.com/p/quanta-bits-july-5-2026
Subscribe to Quanta Bits: https://quanta-bits-newsletter.beehiiv.com/subscribe

Transcript

SPEAKER_00 0:00

For a long time, if you work in IT, a big part of the job was just saying yes. Somebody on the business side found the tool they liked. It usually had a champion, and by the time the request landed on your desk, the decision was basically already made. Your job was to make it happen. And honestly, that was survivable for years. Because when someone bought a bad tool, the damage stayed inside that tool. If marketing picked a bad platform, marketing lived with a bad platform. Nobody else really felt it. The reason that worked is the same reason it stopped working. Hey, I'm Reza Markabody. Welcome back to ContoBids. Last week I talked about how agents are starting to change where work begins. This week is kind of the follow-up to that. Once you let agents in, the way you say yes to new requests has to change too. That's what I want to walk through. So here is the core of it. A normal software tool stays in its own box. It has its own login, its own data, its own walls. If it breaks, it breaks in one corner. An AI request is not like that. When someone asks for an agent or a way to plug one system into another, or access to company data, they're not really asking for one tool in one corner. They're asking for something that can reach across a lot of systems at once. The customer system, where all the sales data lives, the ticketing system, whatever it can get to. So the damage used to stay in one place, now it can touch the whole company. And that changes the math on saying yes. When you say yes to everything without a real process, three things tend to break at the same time. The first is ownership. Nobody actually owns the thing. There was a report from VentureBeat this week making exactly this point. They found that in a lot of companies, no single person is on the hook for an AI project. It just sort of exists. And when something goes wrong, everybody looks around the room. The second is measurement. You can't tell if it worked. A big survey of IT leaders this year found that most of them have committees and approval steps on paper, but only about one in five could say their AI projects actually hit the goals they were after. So the paperwork exists, the results don't. And the third is cost. A regular software license is a fixed predictable bill. AI is different because the cost goes up the more you use it. There was a story about Uber burning through its whole AI budget for the year by April. So the bill doesn't sit still, it grows with use. And here's the thing: none of these are technology problems. They're all decisions nobody made. So what do you actually do about it? I've been working on a version of this with clients and it's honestly not that complicated. It's a handful of moves. 1. Everything comes through the same door. A software request, an agent request, a data request. They all ask the same basic question. So treat them the same way. And renewals count too, because these days AI shows up inside tools you already pay for. 2. The people who own the systems and the people who build the AI sit at the same table, not in two separate silos. They answer simple questions together. Do we already have something that does this? Can we build it? Do we even have the time? 3. You do a real cost check before you buy. And I'll be honest, the hardest part here is that the numbers people bring you are usually way too optimistic. Sometimes that's the vendor selling. Sometimes it's just a team that really wants a yes. So if nobody can actually say what this thing will cost to run, then you're not buying it yet. You're running a small experiment to find out. And four, most requests are small, so give them a fast lane, days, not weeks. You save the full review for the stuff that touches real data, customer records, or real money. Let me make that concrete because it sounds more formal than it feels. Say somebody's sales operations asks for an AI agent that listens to sales calls and updates the customer records afterward. Sounds great. Saves everybody time. Under the old reflex, that's a yes by Friday because the tool has a champion and a demo that looks amazing. Under this process, a few questions come first. Who else does this touch? Well, it writes into the customer system, so the team that owns that system needs to be in the room. Do we already have something close? Maybe the call recording tool the company already pays for just ship this exact feature. That happens constantly now. AI shows up inside tools you already own. And what does it cost to run? If the answer is we are not sure, it depends on usage, then you don't sign a contract. You run it with five sales reps for a month and look at the actual build. Maybe it still ends in a yes. A lot of these should, but now it is a yes with an owner, a way to measure it, and a build somebody has seen. 10 minutes of questions just saves you a year of paying for something nobody owns. And here's where I land on all of this. The process isn't there to say yes faster. It's there to make saying a no possible. Because a no is what protects the yes that actually matters. A couple of other things from this week worth a minute. The first one is kind of wild. One of the most capable AI models on the market, Anthropics Fable, got switched off by a government order for about three weeks and then switched back down. And the interesting part is what happened in between. Most companies had already set up a backup model before the ban even lifted. So the lesson for anyone running this stuff, which model you can use is now partly a policy question, not just a vendor question. Have a backup ready before you need it. The second one is about where the money is actually landing. The consulting firm BCG puts out a big ranking every year of which industries created the most value for their shareholders, and for the first time in over a decade, software way down while things like mining, defense, and banking climbed. The reason is that all the AI money right now is going into the physical stuff, chips, power, data centers, not the software on top. But the same report had a comforting note buried in it. In almost every industry, the best run companies still beat the pack. So a hot industry doesn't save you and a cold one doesn't sink you. How you run the company still matters more than what industry you are in. And one quick thing that has nothing to do with AI just because I thought it was fun. There was a study on plants that I can't stop thinking about. Researchers put a few kinds of barley in separate chambers connected only by air. No roots touching, no light signals, just the air moving between them. And it turns out the plants could tell who their neighbors were just from that. The slower growing ones, when they got air from fast, healthy neighbors, sped up, like they sensed they were about to get shaded out and decided to compete harder. And the fast growers did the opposite. They eased off and put their energy into making their leaves taste bad to bugs instead. Basically, they read the room and picked a strategy, push for growth or dig in and protect what you got. Which honestly is a better prioritization instinct than a lot of companies have. On the personal side this week, I read a book called Plastic Ink by Beth Gardiner. It's a piece of journalism about the plastics industry. I liked it more for the questions it raised than for how airtight it was. A lot of it is stories rather than hard data, but the economics part stuck with me. Her point is that plastic looks cheap only because someone else picks up the real cost. The cleanup, the disposal, the long-term damage. That gets paid by the public, not by the company that paid it. And that question follows you out of the book. Who ends up owning a cost when the people creating it and the people paying for it aren't the same people? A price that looks cheap usually just means someone else is paying it where you can't see. Which, if you think about it, is not that far off from the AI conversation we just had. That's it for this week. The full issue with all the links and sources is in your inbox or at quantabits newsletter.behive.com. I'm Reza Markabadi. Thanks for listening. See you next week.