AI Companies Are Making It Harder to Buy Their Own Products

The models keep getting smarter. The buyers keep getting more lost. Somewhere between the two, the AI industry has a growing business problem.

Gizmodo’s Webb Wright published a sharp piece this week that compares today’s AI industry to the NFT boom at its silliest. His reference point is a May 2022 tweet from NFT marketplace RareCandy. It explained, with a straight face, how “slurp juices” could turn one Bored Ape into several new ones. The tweet opened with “a lotta yall still dont get it.” Most of us never did.

Wright thinks AI now speaks its own version of that dialect. He has a point, and it stings. But the funny names are only the tip of the iceberg. The bigger issue is what happens when an industry that needs everyone builds products that only insiders can follow.

Why the NFT Comparison Stings

NFTs fell apart for plenty of reasons. One of them gets less attention than it should: outsiders couldn’t follow the vocabulary, so they stopped trying. The jargon worked like a velvet rope. Insiders stayed in and everyone else drifted off.

To be fair, AI is not NFTs. People use it every day, pay for it, and get real work out of it. But it has picked up one bad NFT habit. It mostly talks to itself.

The Menu Has Outgrown the Meal

Gizmodo’s list of OpenAI launches since September 2025 says it all:

At its developer conference last month, OpenAI rolled out more than twenty products in a single day. Each one probably makes sense to the team that built it. Lined up side by side, they read like a parts catalog nobody asked for.

No single launch is the problem. The pile is. Before a buyer gets any work done, they have to answer four questions:

• Which model tier fits the task?
• Which product wraps that model?
• Which plan unlocks it?
• How much usage does that plan actually include this month?

Four decisions just to get started. That’s a tough sell to anyone who isn’t already a fan.

Pricing Is Where Confusion Turns Into Distrust

Strange names are annoying. Limits that keep shifting are worse, because they eat away at trust.

Anthropic is fighting a class action lawsuit that claims it misled customers about token allowances on its Claude 5x and 20x plans. Those are allegations, and no court has ruled on them. OpenAI took heat of its own after announcing it would halve the usage allowance on its $200-a-month Pro plan. The loudest complaints came from power users, the very people who usually sing these tools’ praises.

Gizmodo also flags a quieter trend. Some people and businesses are dropping ChatGPT and Claude subscriptions for open-source Chinese models, privacy concerns and all. When customers accept a privacy risk just to get away from your pricing page, the pricing page has a problem.

The timing couldn’t be worse. Public trust in AI was already slipping before these pricing fights started. Every surprise cap adds fuel to the fire.

Here’s the part labs keep missing. Subscribers will forgive a high price. What they won’t forgive is having the rug pulled out from under them mid-month.

The Agent Paradox

For anyone building or buying AI agents, this is where it gets interesting.

Agents are supposed to hide complexity. You state a goal, and the agent handles the steps, the tools, and the small decisions. OpenAI’s dots push that idea the furthest so far. They’re always-on agents that keep working after you close the chat.

Yet buying an agent has become the most complicated part of using one. A team often has to choose:

1. a base model and its variant
2. an agent framework or product
3. an orchestration or decision layer
4. a pricing plan whose limits might change next quarter

So the industry sells simplicity through one of the messiest buying journeys in software. That can’t last. If the agent hides the complexity but the contract puts it all back on the table, customers will judge the product by the contract.

Why the Sprawl Keeps Growing

None of this happens because labs are careless. The incentives push them toward it:

• Launches make headlines. A new model name gets a news cycle. Merging two products gets nothing.
• Throwing spaghetti at the wall. Labs still don’t know what people will reliably pay for, so they ship a lot and see what sticks.
• Copying the neighbors. TechCrunch reported that OpenAI’s Decisions API closely resembles TypeSafe’s Jev. Gizmodo says dots arrived as a direct answer to Meta’s Muse agent. Matching a rival adds another product to the lineup without making any of it clearer.
• Engineers name things. Version numbers and codenames help internal teams. To everyone else, they get lost in translation.

Wright makes a fair historical point here. New categories usually go through a messy stage. Before the iPhone, phones folded, slid, swiveled, and sprouted tiny keyboards. A messy phase is normal. Getting stuck in it is the real danger.

The $6 Trillion Reason This Matters

Then there’s the money. A Bain & Company report estimates the AI industry will need $6 trillion in annual revenue by 2031 just to cover its compute costs.

That pressure is built in. Every query costs real money to serve, so AI doesn’t scale like traditional software. Labs need volume, and volume means mainstream buyers.

Power users alone won’t get anywhere near $6 trillion. The industry needs teachers, accountants, shop owners, and mid-sized firms without an AI team. Those buyers won’t stop to decode “GPT-6 Sol vs. Luna.” They’ll pick whatever feels safe and predictable, or they’ll sit it out.

That’s why clarity is more than a design detail. It’s how these companies pay their bills.

Signs the Industry Is Listening

There’s at least some self-awareness. Thibault Sottiaux, OpenAI’s head of ChatGPT and Codex, posted on X that customer feedback clearly shows people want things simpler, and that the company is on it.

That’s a start. But talk is cheap. The proof will be whether OpenAI and its rivals actually retire products, not just rename them.

What “Simpler” Should Look Like

If AI companies want to shake off the slurp-juice label, here’s where to start:

• Fewer front doors. One product per job, not one per model.
• Names that describe the job. Call it “Research” or “Code,” not an internal codename.
• Limits that hold still. Lock usage allowances for each billing period, and give real notice before changing them.
• Changelogs in plain English. Tell users what changed for them, not what changed under the hood.
• Retirement dates. Publish when old products will go away, so the lineup gets shorter over time.

Less really is more here.

The Bottom Line

AI doesn’t have an intelligence problem. Models are improving faster than most customers can keep up. It has a legibility problem.

NFTs faded partly because only insiders could follow them. AI offers far more real value than NFTs ever did. But value nobody can follow is value nobody pays for. The next edge in AI may not belong to the smartest model. It may belong to the first company whose lineup a regular person can understand in thirty seconds.

Related: The AI Boom Is Minting Billions — But Not Jobs