You Tried ChatGPT Once. That Doesn't Count Anymore.
A founder walked away for four hours. The AI finished the job, tested it, and fixed what it didn't like.
Image by Nano Banana Pro
The Wave Is Here and Most People Are Still Checking the Weather
TL;DR: In the last two weeks, a viral post from an AI founder, a bombshell prediction from Microsoft’s AI chief, new Chinese models matching Western leaders, and a Google benchmark score that shouldn’t exist yet all point to the same conclusion: the pace of AI change just broke away from anything we’ve seen before. This isn’t a tech story anymore. It’s a story about jobs, education, business models, and what you do about it starting now. Here’s what’s happening, why it matters, and what to do whether you’re a parent, a professional, or someone running a team.
“This Seems Overblown”
Matt Shumer builds AI companies for a living. He’s spent six years in the space. And last week, he wrote a post called “Something Big Is Happening” that’s been shared hundreds of thousands of times. The opening line compares where we are right now to February 2020, a few weeks before COVID changed everything.
His point wasn’t about a virus. It was about the gap between what insiders are seeing and what most people believe.
Here’s the part that caught me. Shumer describes telling an AI what to build, walking away from his computer for four hours, and coming back to find the work done. Not a rough draft. The finished product. The AI opened the app, tested the buttons, fixed what it didn’t like, and only came back when it was satisfied with the result. That was his Monday.
I’ve been doing AI consulting for law firms and professional services organizations for years now. I’ve presented AI overviews and live demos now to over 2,500 people. And I can tell you the reaction in the room has shifted. A year ago, people were curious. Six months ago, they were impressed. Now? They’re quiet. The kind of quiet that means someone’s doing math in their head about their own job.
Three Things That Happened in Two Weeks
Let me lay out what actually happened between late January and today, because when you see it all together, the picture gets very clear very fast.
First, the models took another leap. On February 5th, OpenAI released GPT-5.3 Codex and Anthropic released Opus 4.6 on the same day. Shumer called it the moment he realized AI could do his job better than he could. But here’s the detail that stops me cold: OpenAI’s technical documentation says GPT-5.3 Codex was “instrumental in creating itself.” The AI helped debug its own training, manage its own deployment, and evaluate its own test results. That’s not a prediction about the future. That’s a press release about last month.
Second, the people building this technology started saying the quiet part out loud. Mustafa Suleyman, the head of AI at Microsoft, told the Financial Times this week that most white-collar work sitting behind a computer, whether you’re a lawyer, accountant, project manager, or marketing professional, will be “fully automated by AI within the next 12 to 18 months.” Not five years. Not “eventually.” Twelve to eighteen months. And Dario Amodei, the CEO of Anthropic, has been saying AI could displace 50% of entry-level white-collar jobs in one to five years while simultaneously warning that AI more capable than everyone may arrive in just one to two years.
These aren’t bloggers. These are the people spending billions of dollars building the systems.
Third, the rest of the world caught up faster than anyone expected. This week, Chinese AI labs released GLM-5 (745 billion parameters, a rough measure of a model’s complexity) and MiniMax 2.5 (230 billion parameters). Both are open-source. Both are benchmarking near the level of Opus 4.6 in coding and reasoning tasks, at a fraction of the cost. And then Google dropped a major update to Gemini 3 Deep Think that scored 84.6% on a notoriously difficult reasoning test called ARC-AGI-2, designed to measure the kind of abstract reasoning that AI models have historically struggled with. For context, Claude Opus 4.6 scored 68.8% on the same test. GPT-5.2 scored 52.9%. Gemini 3 Pro Preview scored 31.1% Google improved 2.5x against its prior best model and beat the current leader (Opus) by 23%.
When I wrote about the legal tech crash two weeks ago, the $300 billion market wipeout after Anthropic released its legal plugins, I said the real story wasn’t the tool. It was the signal. Foundation model companies (the ones building the core AI engines like Claude, ChatGPT, and Gemini) are now building industry-specific products that compete with their own customers. These three developments in two weeks? Same pattern, bigger signal. The technology isn’t just improving. It’s improving from multiple directions simultaneously, and the timeline for “when this affects me” just got a lot shorter.
And Then a Lobster Showed Up
While the big labs were leapfrogging each other, something happened at the individual level that might matter just as much.
An Austrian developer named Peter Steinberger released an open-source personal AI agent originally called Clawdbot (named after the little monster you see when Claude is loading). After a trademark complaint from Anthropic and a couple of name changes, it became OpenClaw. Within weeks, it had over 145,000 developer endorsements on GitHub (the platform where open-source code lives) and millions of installs.
Here’s why it matters. OpenClaw isn’t a chatbot. It’s an agent. You install it on your own machine, connect it to a large language model, and then interact with it through WhatsApp, Telegram, Signal, or Slack. It reads your email. It updates your calendar. It browses the web on your behalf. It summarizes documents. It remembers your preferences across sessions and adapts to how you work. One user described it as “AI with hands.” Another called it “Jarvis, and it already exists.”
I got my own OpenClaw instance running last week, and the value is already real. It’s handling scheduling, research summaries, and routine communications that used to eat an hour of my morning. It’s early, it requires some technical comfort to set up, and security researchers have raised valid concerns about the broad permissions it needs. But the trajectory is clear: personal AI agents that actually do things, not just answer questions, are here now. Not in a research paper. On your phone.
People called ChatGPT the “iPhone moment” for AI. OpenClaw might be the moment AI agents crossed the same line. Zacks called it “agentic AI’s ChatGPT moment.” And the fact that it’s open-source, free, and built by one developer using existing AI models tells you everything about how fast this space moves. A single person with a laptop just built something that rivals what billion-dollar companies have been promising for years.
Why This Time Is Different (And Why “I Tried ChatGPT” Doesn’t Count)
I hear a version of this every week: “I tried it. It wasn’t that good.”
I get it. If you used ChatGPT in 2023 or early 2024, you were probably underwhelmed. It made things up. It got confident about wrong answers. It felt like a clever parlor trick.
That was a different era. Shumer put it well: judging AI based on the free tier of ChatGPT is like evaluating the state of smartphones by using a flip phone. The gap between the free versions and what paying users have access to is now enormous. And the gap between what was available six months ago and what dropped this month is even bigger.
I see this in my own client work. Last month I ran a live demo for a family law firm. We took a real custody case and had Claude work through the research, draft a memo, and pull relevant precedents. Work that would have taken a first-year associate most of a day. The AI did a credible first pass in about eight minutes. The partner in the room got very quiet, then asked the question I keep hearing: “So what do we hire the new people to do?”
That’s not a technology question. That’s a business model question. And it applies to a lot more than law.
The Part Nobody Wants to Talk About
Here’s what keeps me up at night, and it’s not about AI replacing jobs. It’s about the training pipeline.
Think about how every prestigious career actually works. You start at the bottom doing the repetitive stuff. Reviewing contracts. Building financial models. Assembling pitch decks at 2 a.m. It’s tedious. Sometimes it’s mind-numbing. But that grunt work is stealth training. You learn to spot the thing that doesn’t fit. You develop pattern recognition. You build the judgment that eventually makes you valuable enough to run the team.
AI is eating that bottom rung first.
I’ve written about this before as the pyramid-to-diamond shift. The wide base of entry-level workers doing grunt work is shrinking fast. Companies are hiring fewer people out of school. The ones they hire start higher up, managing AI tools instead of doing the repetitive work themselves. The middle of the organization swells with experienced people who can direct AI to be incredibly productive. The top stays small.
Sounds efficient. And it is. But here’s the problem: if the junior people never do the grunt work, how do they develop the judgment that made you good at your job? If AI handles the document review, the data pulls, the first-draft memos, we save money today. But we create a gap in the pipeline that shows up five or ten years from now.
This is a succession crisis hiding inside an AI adoption story.
What This Means If You Have Kids in School
I have three kids navigating this. A senior in college, a junior, and a high schooler watching it all unfold. The advice I give them is different from the advice I was given, because the game changed under our feet.
Most colleges are still stuck in the “is this cheating?” phase of AI. Meanwhile, the job market has already moved to “this is the baseline.” A marketing director told me recently that she assumes every applicant knows how to use AI. What she’s hiring for is the person who knows when the AI is wrong.
That’s a different skill. And almost no curriculum is designed to build it.
The old deal was: work hard, follow the path, get the degree, and the system takes care of you. The new deal is: build the skills the system can’t automate, because the system is being rebuilt around you. That means judgment. Emotional intelligence. The ability to frame a problem worth solving, direct tools toward it, verify the output, and make a call when the answer isn’t clean.
Those skills aren’t on most syllabi. And they’re about to become the only ones that matter.
The Uncomfortable Math for Professional Services
If you run a law firm, an accounting practice, or any professional services business, the revenue model question is more urgent than the tool question.
KPMG recently argued for, and got, a 14% reduction in audit fees by pointing to AI efficiencies. Baker McKenzie laid off staff, citing AI-driven changes in work volume. These aren’t hypothetical scenarios. They’re happening now.
The traditional model of selling time for money assumes that the value of work correlates with the time spent producing it. AI breaks that assumption permanently. When contract review takes three minutes instead of four hours, you can’t bill 0.1 hours for what used to be a $2,000 task. And you can’t pretend the work still takes four hours when your client’s general counsel just read the same headlines you did.
But here’s the part that should actually excite you: for firms that move early, AI doesn’t destroy margins. It can transform them. If you can deliver a $5,000-value contract review for $2,000 using AI-assisted workflows, and your cost drops from $1,800 in associate time to $200 in AI tooling plus partner review, your margins don’t shrink. They grow. But only if you price on value, not on hours.
The firms that survive this aren’t the ones that pick the right AI tool. There is no right tool. The tools are changing too fast. The firms that win are the ones that redesign their business around what AI makes possible.
What To Do Now
I’ve organized this by audience because the actions are different depending on where you sit. But they all start from the same place: stop waiting and start using AI daily.
If you’re an individual or early in your career: Get a paid subscription to at least one AI tool (Claude, ChatGPT, or Gemini) and use it for real work every day. Not as a toy. As a colleague. Learn where it’s strong and where it falls apart. Build the habit of checking AI output against your own judgment. That verification skill, knowing when the output is wrong, is rapidly becoming the most valuable thing you can bring to any job.
If you’re a parent or educator: Push your school to integrate AI into the curriculum as a tool, not treat it as a threat. Ask colleges how they’re building judgment and verification skills, not just testing recall. Encourage projects where students use AI and then document what they changed, what they caught, and why. That process of showing your thinking is the new “showing your work.”
If you’re in professional services: Run the numbers on what happens when 30% of associate-level work gets automated. Pick one practice area and pilot value-based pricing alongside your hourly billing. Start the talent conversation now, because the apprenticeship model that trained you is breaking, and whoever figures out a new development pathway first has a real competitive advantage.
If you’re a business leader: Audit your vendor stack for middleman risk, meaning tools that are thin layers on top of the core AI engines and could be disrupted overnight, the way legal tech stocks just were. Set a 90-day goal to have every knowledge worker on your team using AI daily for real tasks. And start thinking about your business model, not just your tools. The technology will keep changing. Your pricing, your service design, and your talent pipeline need to change with it.
The Gap Is the Danger
The biggest risk right now isn’t AI itself. It’s the gap between what insiders know and what most people believe.
Shumer compared it to February 2020. I think he’s right about the dynamic, if not the exact analogy. The people closest to this technology are sounding alarms, and most of the world is still treating it like an interesting dinner-party topic.
The models are improving from every direction. American labs, Chinese labs, open-source, closed-source. The benchmarks that were supposed to take years to crack are falling in months. The CEOs building these systems are putting 12-to-18-month timelines on changes that will reshape entire industries.
You don’t have to believe every prediction. I don’t. Timelines slip. Regulation intervenes. Reality is messier than forecasts.
But the direction is clear. And the cost of waiting to see how it plays out is going up every month.
The best time to start was six months ago. The second-best time is today. Open the tool. Do the work. Build the skill. Because the wave isn’t coming. It’s here.
In this article, we looked at the gap between what AI insiders are seeing and what most professionals, parents, and business leaders still believe. Four major developments landed in two weeks. The models got better from every direction. The people building them put short timelines on massive disruption. And personal AI agents crossed from research project to daily tool. The signal is loud, but the path forward is simpler than the noise suggests.
If you want help sorting this out:
Reply to this or email me at steve@intelligencebyintent.com. Tell me where you or your team is stuck, what’s slowing your workflow down, or what questions you’re trying to answer about AI adoption. I’ll tell you what I’d test first, which part of the current AI stack (Claude, ChatGPT, Gemini, or the new open-source agents) fits your situation, and whether it makes sense for us to go further than that first conversation.
Not ready to talk yet?
Subscribe to my daily newsletter at smithstephen.com. I publish short, practical takes on AI for business leaders and professionals who need signal, not noise.