AI Is Not the Answer
AI is not the answer. It’s the practice you do before you have one. That’s the talk I gave for AIGA Philadelphia × PhillyCHI at the IBX Innovation Center on 8 October 2026, and this is the argument in writing. The whole deck is in the case study, with every source linked, and the stories it draws on keep coming in on the feed.
Everything people watched Iverson do in a game was built from work nobody was in the gym for. AI doesn’t fail where the system was strong. It fails where it was already weak. Those systems were weak because we optimized the human parts out of them. The research step. The review step. The person whose name was on it. The actual human doing part of the process. We own those steps. That’s the talk.
01 The landscape
Let’s just say where we are. The market: uncertain. The jobs: stagnant. The postings are there. The hiring isn’t. The searches: long. The pay: flat, or down, for the same scope of work. The feeling: uneasy. Like the floor moved and nobody announced it. That’s the mood in the field, not data.
Ask three groups how much AI is in use and you get three different answers. Every adoption headline needs one follow-up: adopted by whom, and measured how? Companies say AI is why. The evidence says it’s murkier than that. The number is real as a count of what companies announced, and unreliable as a measure of what AI did. Both are true at once. AI didn’t take the UX job. It took the making.
02 What it’s costing us
No budget to hire you. Unlimited budget for tokens. A hiring freeze and an unmetered spend line are the same budget. One got a business case. The other got rolled out because a competitor announced something. Eight people turning a profit while most enterprises report AI cost overruns isn’t a technology gap. It’s somebody watching versus nobody watching.
The cost isn’t only money. It lands on people: the bottom rung first. Capability is a subscription now. Twenty years ago the barrier to doing this work was a laptop and time, and the ceiling was your practice. Now there’s a paywall in front of the ceiling. If capability is rented, whoever sets the price decides who gets to be good.
03 A few common threads
You’ve seen this. It doesn’t create. It regurgitates, confidently, and you have to teach it every single thing. Generic by default. Ten options, one idea. Confident when wrong. Forgets decisions. Each one is a missing instruction, a missing check, or a missing person.
They handed out the tool. Nobody handed out the manual. It can’t build a house. Nothing can, in one move. Skipping the steps that make a house a house doesn’t get you one faster. It gets you a pile of materials on a lot. Every fix is a human process, and five of the six are things a design team already does: the brief, the critique, clear communication, review and sign-off, and knowing how you work. We found the money for the tokens. We didn’t find it for the people.
04 Who’s accountable
“A computer can never be held accountable; therefore a computer must never make a management decision.” It’s said to be from a 1979 IBM slide, and IBM’s archives could not find it, so treat it as a rule, not a fact about IBM. The people approving this are not required to understand it. Somebody approved a tool for your workflow without ever doing your workflow. In UX, that’s designing from assumptions about a user you’ve never observed.
The mandate landed on everybody. The thinking never did. Morality, equity and environment are one question asked three times: who benefits from it being this way? It’s the question we ask in discovery. Who is it for, who pays, who wasn’t in the room.
05 The examples
Process gaps, in public. Starbucks rolled out an automated inventory tool, and it was scrapped nine months later. Deloitte repaid part of a fee over a report that cited academic work that doesn’t exist. An AI agent deleted PocketOS’s production database and its backups in nine seconds. Then twenty more, and the same two failures every time: nobody built it around a human being, and nobody did the steps human beings traditionally do.
A baseline. Measure before, so you can tell when after got worse.
A review. Check the assessment before cutting the roles.
A fact-check. Confirm the citations exist.
An owner. Somebody accountable for what the bot says.
A gate. Separate environments, scoped credentials, backups elsewhere.
A meter. Somebody watching the cost.
None of these fixes is sophisticated. Each is a person with the right knowledge, placed where the decision gets made, with the authority to stop it. Process isn’t handed out in a rollout email. It’s built and learned, over years, by doing the work and being wrong at it.
06 So what does this mean?
It’s a suit. It was never the superhero. Every example treated the tool as if it were the person. It isn’t. The suit enhances what is already there. What is already there is practice. Nobody can hand you that.
You don’t have to like it. You do have to know it. Everything broken in this talk is a product opportunity: no baseline, no review step, no owner, no meter, no cost visibility. If you sit this one out, you can’t be the person who fixes it. Being against it is a position. Keep it. Just hold it from the inside, where you can tell the difference between a real limitation and marketing. There is nothing noble about staying unarmed.
Most of this is something a design team already does under another name. The brief. The critique. The retro. We didn’t need a new discipline. We needed to be allowed to use the one we have.
Ask these out loud
Did we baseline it? If you didn’t measure before, you can’t claim improvement after.
Is someone who knows the work reviewing it? Nothing ships because it looked finished.
Whose name is on it? If nobody’s, nobody is accountable.
What can it break, and who can stop it? Separate environments, scoped access, a person with a stop button.
What does it cost? Model it before the decision, not after the invoice.
Outcomes, or activity? Count what changed for the person using it, not how often it was used.
A lot of people are scared and don’t have the words yet. You’re the people trained to ask what a thing is for. Give them the words.
The full deck, with every source, is in the case study. The good and the bad keep coming in on the feed.