OpenAI’s GPT-5 is unlikely to offer AI Brokers a lot of an improve

As 2025 dawned, OpenAI CEO Sam Altman was selling two developments he insisted would remodel our lives. One, in fact, was GPT-5 — a long-anticipated main improve to the Massive Language Mannequin (LLM) that powered ChatGPT’s rise to tech world superstardom.

The opposite? AI Brokers that do not simply reply your queries like ChatGPT, however truly get stuff executed for you. “We consider that, in 2025, we may even see the primary AI brokers be a part of the workforce and materially change the output of firms,” Altman wrote again in January.

Nicely, we’re eight months in, and Altman’s prediction already wants a giant previous asterisk. Positive, firms are eager to undertake AI Brokers, akin to OpenAI’s ChatGPT agent. In a Could 2025 report, consultancy big PWC discovered that half of all companies surveyed deliberate to implement some type of AI Agent by the tip of the 12 months. Some 88% of executives wish to enhance their groups’ AI budgets due to Agentic AI.

However what in regards to the precise AI Agent expertise? With apologies to all these hopeful executives, the critiques are nearly uniformly adverse.

If “AI Brokers” was a brand new high-tech James Bond film, here is the type of blurbs you’d see on Rotten Tomatoes: “glitchy … inconsistent” (Wired); “got here off like a clueless web beginner” (Quick Firm); “actuality would not dwell as much as the hype” (Fortune); “not matching as much as the buzzwords” (Bloomberg), “the brand new vaporware … overpromising is worse than ever” (Forbes).

Research finds OpenAI’s entry failed practically each time

A Could 2025 Carnegie Mellon College examine (PDF) discovered Google’s Gemini Professional 2.5 failed at real-world workplace duties 70% of the time. And that was the greatest-performing agent. OpenAI’s entry, powered by GPT 4.o, failed greater than 90% of the time.

GPT-5 is probably going to enhance on that quantity … however that is not saying a lot. And never simply because early reviews say OpenAI struggled to fill GPT-5 with sufficient enhancements to make it worthy of the discharge quantity.

Certainly, it is beginning to look to researchers like this disappointment is baked in to the entire means of LLMs studying to do stuff for you. The issue, as this AI Agent engineer’s evaluation makes clear, is simple arithmetic: errors compound over time, so the extra duties an agent does, the more severe they get. AI Brokers who do a number of complicated duties are liable to hallucination, like all AI.

Ultimately some brokers “panic” and may make “a catastrophic error in judgment,” to cite an apology from a Replit AI Agent that actually deleted a buyer’s database after 9 days of engaged on a coding job. (Replit’s CEO known as the failure “unacceptable”.)

Tellingly, that is not the solely AI-Agent-wipes-code story of 2025 — which explains why one enterprising startup is providing insurance coverage in your AI Agent going haywire, and why Wal-Mart has had to herald 4 “tremendous Brokers” in a bid to corral its AI Brokers.

No surprise a current Gartner paper predicted that 40% of all these AI Brokers at the moment being initiated by firms might be canceled inside 2 years. “Most Agentic AI tasks,” wrote senior analyst Anushree Verma, are “pushed by hype and misapplied … This will blind organizations to the actual price and complexity of deploying AI brokers at scale.”

What can GPT-5 do for AI Brokers?

It is attainable that ChatGPT agent will vault to the highest of the reliability charts as soon as it is powered by GPT-5. (Once more, that is not the best of boundaries.) However the brand new launch is unlikely to repair what actually ails the Agentic world.

That is as a result of guardrails are already being erected — by firms in addition to regulators — shutting down what even probably the most dependable AI Agent can do for you.

Take Amazon, for instance. The world’s largest retailer, like most tech giants, is speaking a giant recreation on AI Brokers (as they did at a Shanghai Agentic AI honest in July, pictured above). On the identical time, Amazon has shut down the power of any AI Agent to browse and purchase wherever on its website.

That is smart for Amazon, which has all the time needed management over the shopper expertise, to not point out its need to ship advertisements and sponsored outcomes to precise human eyeballs. Nevertheless it’s additionally curbing an enormous quantity of potential Agent exercise proper there. (On the plus aspect, no “catastrophic failure” involving a big pile of next-day deliveries at your door.)

And can we belief AI Brokers to purchase on-line for us anyway? It is not that they are evil and wish to steal your bank card knowledge; it is that they are naive and susceptible to being phished by unhealthy actors who do need your card.

Even GPT-5 might not be capable of get round one vulnerability seen by researchers: knowledge embedded in photographs can instruct AI brokers to disclose any bank card information they could have, with the person being none the wiser.

If that type of drawback is exploited on a company scale, then Altman could also be proper about AI Brokers “materially altering output” — simply not in the way in which he meant.