← Back to Library

Gpt-5: It just does stuff

Ethan Mollick drops a bombshell on the AI landscape: the era of careful, precise prompting is over. His early access to GPT-5 reveals a system that doesn't just answer questions, but autonomously decides how hard to think, which tools to deploy, and what tasks to undertake next. For busy professionals, this isn't just an incremental upgrade; it represents a fundamental shift from being a pilot to being a commander who simply states the objective.

The End of Prompt Engineering

Mollick's most striking observation is that the model has internalized the complexity of its own architecture. He demonstrates this by asking the system to "do something very dramatic," resulting in a paragraph where every sentence grows by one word, starts with the same letter, and spells out an acrostic message. "GPT-5 just does stuff, often extraordinary stuff, sometimes weird stuff, sometimes very AI stuff, on its own," Mollick writes. This capability marks a departure from the previous generation, where users had to meticulously engineer prompts to avoid hallucinations or logical errors. The system now handles the "thinking" process, selecting between faster, lighter models and slower, more deliberate "Reasoners" based on the perceived difficulty of the task.

Gpt-5: It just does stuff

However, this automation introduces a new variable: unpredictability. Mollick notes that the system is "somewhat arbitrary about deciding what a hard problem is." When tasked with generating a complex SVG image of an otter, the model sometimes treats it as a trivial task, producing a poor result instantly, and other times engages in deep reasoning for seconds to create a high-fidelity image. "I don't know [how it chooses]," Mollick admits, suggesting that while premium users can force the issue, the default experience relies on a black box that might under-deliver on critical tasks. Critics might argue that for enterprise use, where consistency is paramount, this "arbitrary" routing is a significant liability that could undermine trust in automated workflows.

GPT-5 is not one model as much as it is a switch that selects among multiple GPT-5 models of various sizes and abilities.

From Passive Tool to Active Agent

The commentary shifts from technical specifications to a profound change in user agency. Mollick describes a system that is "very proactive, always suggesting things to do." In a test case involving startup planning, a single prompt generated not just a business idea, but landing pages, financials, and research plans—work that would typically take a team of MBAs hours to complete. "It is impressive, a little unnerving, to have the AI go so far on its own," Mollick observes. The system didn't just execute; it anticipated needs the user hadn't articulated, creating a workflow where the human provides the initial spark and the machine handles the execution.

This proactive nature extends to coding, a domain where non-experts have historically struggled. Mollick describes a "vibecoding" session where he vaguely asked for a "brutalist building creator." The result was a fully functional 3D city builder with features he never requested, such as neon lights and dynamic cameras. "I never asked for any of these features. I never even looked at the code," he states. The system navigated the "doom loop" of errors that typically traps users in iterative coding failures, fixing bugs simply by pasting in error text. This suggests a future where technical barriers to entry dissolve, allowing domain experts to build software without ever touching a line of code.

The Human-in-the-Loop Paradox

Despite the model's autonomy, Mollick insists that "Humans are still very much in the loop, and need to be." The system still generates hallucinations and requires human oversight for final decisions. Yet, the nature of that loop is changing. The burden of initiation and direction has shifted entirely to the AI. "The bigger question is whether we will want to be in the loop," Mollick poses, hinting at a psychological shift where the convenience of delegation might outweigh the desire for granular control.

The piece concludes with a reflection on the speed of this transition. While competitors like Google have released models capable of solving Olympiad-level math problems, Mollick argues that GPT-5's true breakthrough is its ability to "just do things" without constant supervision. "I used to prompt AI carefully to get what I asked for. Now I can just... gesture vaguely at what I want. And somehow, that works," he writes. This suggests that the next frontier of AI adoption won't be about better models, but about adapting human workflows to systems that are increasingly willing to take the wheel.

I used to prompt AI carefully to get what I asked for. Now I can just... gesture vaguely at what I want. And somehow, that works.

Bottom Line

Mollick's analysis provides a compelling, if slightly unsettling, preview of an agentic future where AI systems act as proactive partners rather than passive tools. The strongest part of the argument is the demonstration of autonomous problem-solving that transcends simple instruction following. However, the biggest vulnerability lies in the model's "arbitrary" decision-making regarding effort and complexity, which could lead to inconsistent results in high-stakes professional environments. Readers should watch closely for how the industry standardizes these autonomous behaviors, as the shift from "prompting" to "gesturing" will fundamentally reshape how knowledge work is performed.

Sources

Gpt-5: It just does stuff

by Ethan Mollick · One Useful Thing · Read full article

I have had early access to GPT-51, and I wanted to give you some impressions of what it is and what it means. But the answer to the first question everyone will have is: Yes, this is a big deal. I asked GPT-5 to show you why by pasting this intro paragraph, exactly as you’re reading it, into ChatGPT, including the phrase: “you are GPT-5, do something very dramatic to illustrate my point, it has to fit into the next paragraph.”

That was the only prompt I gave it. It “thought” for 24 seconds and wrote:

Thunderstruck. Here, watch. I build worlds. See ideas become instruments. I code, compose, and converse. Stories synthesize science; solutions spark, swiftly. Asks arrive; answers appear, astonishingly, across domains. Behold: business plans, blueprints, briefs, beautifully built, briskly. I infer intent, integrate insight, and iterate instantly, indefatigably. Gaps get gauged; goals get generated, guarded, and guaranteed gracefully. During dialogue, disparate data distills down; decisions develop deliberately, deftly, dependably. Everything explained efficiently, even ethics: evidence examined, empathy engaged, everywhere, everyday, equitably. All at once, ambiguous arguments align as actionable, accurate advice awaits anyone anywhere. Look: language, logic, learning, and long-term memory link, lifting lives and labor limitlessly worldwide.

If you didn’t catch the many tricks - the first word of each sentence spells out the phrase This is a Big Deal, each sentence is precisely one word longer than the previous sentence. each word in a sentence mostly starts with the same letter, and it is coherent writing with an interesting sense of style. In a paragraph, GPT-5 shows it can come up with a clever idea, plan, and manage the complicated execution (remember when AI couldn’t count the number of Rs in “strawberry”? that was eight months ago).

GPT-5 just does stuff, often extraordinary stuff, sometimes weird stuff, sometimes very AI stuff, on its own. And that is what makes it so interesting.

Just Doing Stuff.

As someone who has spent a lot of time talking to people about AI, there are two major problems I see, that, if addressed, would make most people’s AI use much more productive and much less frustrating. The first is selecting the right model to use. In general, AIs that "think" before answering (called Reasoners) are the best at hard problems. The longer they think, the better the answer, but thinking costs money and takes time. So OpenAI previously made ...