A digital crawfish is moving across your screen. It opens a research portal, downloads a fifty-page industry report, pulls out the key numbers, and drops them neatly into a PowerPoint slide. It doesn’t ask for a lunch break. It doesn’t complain about the font size. It doesn’t even have eyes, yet it sees exactly what it needs.
This little digital “creature” – known as Openclaw – scuttles through tasks that once took a human intern three cups of coffee and an afternoon of quiet despair. In doing so, it makes me wonder: if the crawfish does all of that for me, what exactly am I supposed to learn?
During the summer of my sophomore year, I landed my first internship at a consulting firm tucked away in an industrial park. I was a green hand – an industry research intern who knew absolutely nothing about the New Energy Vehicle sector. Curiosity was my only compass. My days weren’t spent prompting AI; they were spent manually searching websites for public records, hunting down obscure industry reports, and cross-referencing news clips. I was the one building the industry “mapping” from scratch, slide by painstaking slide.
At the time, I tried using the AI tools available, but they were unreliable assistants. They hallucinated sources or produced generic outlines that barely scratched the surface of my needs. Frustrated, I turned to my mentor for help. She didn’t hand me a better software link; instead, she gave me a piece of advice that felt old-school, but sounds like a luxury now: “Take your time,” she said. “Do it by hand. Read it yourself.”
I spent an entire week on a single task. I read every word, organised every data point, and double-checked every calculation. But in that slow, manual grind, something happened that no AI can simulate: I learned. By moving the data myself, I began to understand the basic logic behind the industry. I started to see why certain numbers mattered and how investment thinking is structured.
It was the “First Rung” of my career – not just repetitive labour, but the essential practice required to build a professional mind.
That manual grind now feels like a snapshot of a fading world. I was on one of the last trains leaving the station of the traditional era – when humans were still allowed the time to be “inefficient” enough to actually learn. Today, however, that protective bubble of slow growth has burst. The ladder I climbed is being structurally rebuilt, shifting the focus from how we think to how fast we can execute. This acceleration is driven by a technological leap that moves far beyond conversation.
The leap from ChatGPT to the current era represents a fundamental shift: the transition from ‘language’ to ‘action’. Leading this charge is Openclaw, an open-source autonomous AI agent platform with the icon of a crawfish, which acts as a “digital employee.” Unlike standard chatbots, it operates as an agentic interface capable of executing complex workflows across local machines and web browsers. By integrating with Large Language Models – LLMs – like Claude or GPT-4, it can manipulate cursors, manage files, and automate multi-step professional tasks directly through interfaces like WhatsApp or Slack.
This isn’t just a theoretical tool for hobbyists, it is already being used in top-tier firms. My friend Silver, a postgraduate student who recently secured a job offer after multiple high-profile internships, showed me the power of internal financial AI agents. His AI can automatically interpret individual stock news, correlate price fluctuations, and answer complex investment queries. By inputting a target company, a professional-grade investment report is generated in seconds.
Looking back at my internship, there was a moment of frustration that I now see in a different light. I had poured my heart into a research draft, only to have my mentor tell me it didn’t meet the “professional level.” Because I was slow and the quality was poor, she eventually pivoted: she gave me the “easier” tasks – basic data cross-checking and bibliography formatting – while she wrote the core analysis herself.
At the time, I felt relieved. My workload was lighter. But after my departure a year later, a thought struck me: if an AI like Openclaw had been there to do those “easy” tasks for me, would I have actually learned anything? Or would I have just moved through the internship without ever understanding investment research?
In finance, the growth ladder has long followed a predictable trajectory: from manual execution to understanding, and finally to independent judgement. The “grunt work” – running financial models, formatting pitchbooks, reconciling spreadsheets – was the soil in which the seeds of a senior analyst were planted. Yet when AI swallows these hands-on tasks, it removes the very nutrients required for professional growth.
This is the essence of the First Rung Crisis: a structural collapse of the entry-level experience. For decades, the “boring” work was the silent tutor of the professional world. By handing these foundations over to digital agents, we are not just saving time; we are sawing off the very rungs required to climb toward expertise. A 2025 theoretical economic model published by Ide estimates that AI-driven automation of entry-level tasks could reduce US long-run economic growth by 0.05 to 0.35 percentage points annually – not through unemployment, but through a slow erosion of the skills pipeline that organisations depend on to handle complex work.
We are witnessing a dangerous “jump” in professional development. The industry is asking the next generation to become supervisors before they have ever been workers, and to exercise judgement before they have built intuition. This crisis isn’t about a lack of jobs; it’s about the disappearance of the friction required to grow.
The “First Rung” isn’t just about output; it’s about cognitive friction. In the manual era, when a junior intern spent a week reading every line of a 50-page prospectus, that “slowness” was a feature, not a bug. The friction of the work forced my brain to synthesise information, spot contradictions, and develop a “feel” for the data.
However, with Openclaw and other agents, we can achieve “Zero Friction.” You input a command in human language, and the result appears. But without the friction of doing, there is no heat of learning. We are producing a generation of “supervisors” who have the authority to check the AI’s work, but lack the intuition to know when that work is “hallucinating”.
AI has lowered the barrier to entry for many roles. One no longer needs to master a vast amount of knowledge in a specific field; instead, by using natural language to input prompts, one can extract decent results from massive amounts of literature. Meanwhile, as our experience with AI over the past few years has shown, AI is not always correct. Its outputs require human verification and intervention. Yet, those lacking professional depth can only perform basic checks – like verifying information sources – rather than retracing and verifying the logical thought process the AI used to complete the task.
This leads to the “Black Box Supervisor Dilemma”: you are managing a process you cannot replicate by hand. Once the AI system fails, the new generation of analysts will be intellectually paralysed.
The real danger of the “First Rung Crisis” is like always using a calculator for basic maths. If you never practise adding and subtracting in your head, your “maths muscle” will shrink. If young people only use AI to do the “hard thinking” or “organising,” they are outsourcing their brains. When a “Black Swan” event happens – a giant surprise that wasn’t in the history books – the AI will fail, and the humans who forgot how to think for themselves won’t know how to take over the controls. We are trading our ability to handle “surprises” for the convenience of “speed.”
The job market is already reacting. In my recent job search, the shift was stark. Leading firms like Temasek are no longer looking for “Office experts”; they are hiring “Generative AI Citizen Developer Intern.” The Job Description has evolved from “Strong analytical skills and attention to detail” to “Basic understanding of Generative AI prompt engineering and various prompt engineering techniques.”
According to the 2024 Work Trend Index by Microsoft and LinkedIn, 66% of leaders say they would not hire someone without AI skills. “Survival of the Fit” here means that those who master AI agents can complete the work of ten interns with a single click. The intern is no longer a “prospective labourer” learning the craft; they must immediately become an “AI Supervisor.” This reshuffles the hierarchy: only those who can bridge the gap between AI automation and high-level judgement will survive the pruning of the professional ladder.
But let’s be clear: this “Survival of the Fit” is a double-edged sword. To “fit” into the AI era, you must be more than a prompt-user. You must be a deep thinker who uses AI to handle the mundane, while intentionally seeking out manual “deep dives” to maintain your edge. The real “Fit” aren’t those who use AI the most; they are those who know when to turn it off.
The tragedy of the First Rung Crisis isn’t that the machines are taking our jobs; it’s that they are taking our path to expertise.
However, for cost-conscious employers, the expense of hiring these “digital employees” is significantly lower than that of an intern. My summer internship salary was approximately $600 per month. According to Hostinger’s 2026 AI cost breakdown, running OpenClaw (as an open-source tool) costs between $50 and $100 per month in AI tokens and server rentals. Beyond the stark cost advantage lies an efficiency gap: when the McKinsey Global Institute notes that AI can automate 60% to 70% of repetitive tasks in banking, “manual labour” ceases to be a training ground and turns into a “cost centre.” For a firm, paying an intern to manually collect data that an AI can fetch in seconds is not just inefficient – it is a betrayal of business logic.
As Silver demonstrated with his Investment Agent, a newcomer can now replicate the logic of investors like Warren Buffett with a single click. From this perspective, AI is the ultimate equaliser. It bridges the gap between a sophomore and a senior analyst, allowing the human to focus on what their Industry Dashboard shows: high-level visualisation, strategic storytelling, and immediate insight. The “Fit” who survive are those who stop acting like a “search engine” and start acting like a “System Architect.”
Digital agents offer immense advantages in efficiency and cost if you only care about the result. But as we transition from “Learning by Doing” to “Learning by Supervising,” we encounter a problem: how can you supervise a process you don’t understand?
We must address the fundamental flaw in AI’s architecture: it is a machine of induction. AI systems, including the most advanced agents, build their worldviews by analysing millions of historical data points to predict the next “most likely” outcome. In a 2025 ACL study, Li and colleagues systematically evaluated LLMs’ capacity for inductive reasoning – the ability to infer general rules from specific observations. They found that even minor noise in input data causes LLMs to suffer from “hypothesis drift” and “pattern overfitting,” with models consistently defaulting to memorised patterns rather than forming stable, abstract generalisations. It is, by definition, backward-looking.
Imagine you are playing a video game. An AI bot has played this game a million times and knows every hidden path. But suddenly, the developers release a massive update that changes the gravity and the map entirely. The AI bot will keep running into walls because it only knows the “old map.” As a PNAS study on ‘gray swan’ tropical cyclones demonstrated, AI weather models trained on historical data cannot extrapolate to predict Category 5 storms if similar events were absent from their training set. The models do not understand the underlying physics; they simply recognise patterns. When a pattern never before seen emerges – a 200-year flood, a pandemic-driven supply shock, a sudden shift in monetary policy – the inductive machine has no foundation from which to reason.
This is what happened in 2022. The world faced things that hadn’t happened in decades – prices of everything suddenly doubling (inflation) and global tensions we hadn’t seen before. The AI, looking at the “old map” of the last 10 years, couldn’t understand it. Only a human could look at the new situation and say, “Wait, the rules have changed, we need a new plan.” This is deduction – thinking from scratch.
The problem with moving straight to a “Supervisor” role is the loss of tacit knowledge – the kind of wisdom that isn’t written in a manual but is absorbed through hours of manual work. When you manually clean a messy Excel sheet, you aren’t just moving numbers; you are learning the “smell” of the data. You begin to notice that a certain company’s margins always look a bit too perfect, or that a specific line item in a balance sheet feels “heavy.”
AI gives you the result, but it strips away the context. As a supervisor who has never done the work, you can check if the AI’s output matches the template, but you will never notice the “silent red flags” that only reveal themselves when you are knee-deep in the raw, unpolished facts.
The danger isn’t that AI will make a mistake; the danger is that it will produce a result so ‘perfect-looking’ that we no longer have the capacity to recognise the mistake even if it’s staring us in the face.
In the pre-AI era, the “First Rung” was a low-stakes sandbox. We made small, embarrassing mistakes – formatting errors, minor calculation flips – and were corrected by mentors. These “small fires” taught us how to prevent total meltdowns later in our careers.
Now, because AI produces a result that looks “95% professional” instantly, the learning-by-failure loop is broken. The intern is under immense pressure to be “perfectly right” because the AI already provided the “almost right” version. There is no longer a space to be a “clumsy beginner.” This creates a generation of professionals who are fragile: they have never learned how to recover from a logic error they made themselves, because they only know how to polish an error the AI made for them.
If every firm uses AI agent to replicate the logic of Warren Buffett, eventually every investment report will look the same. We are heading toward a commoditisation of insight.
The “First Rung Crisis” doesn’t just hurt the individual; it hurts the industry. When everyone outsources the “heavy lifting” to the same set of algorithms, the “Alpha” – the unique edge – disappears. True mastery comes from the weird, manual deep-dives that AI wouldn’t think to do. By automating the first rung, we are effectively pruning the “wild branches” of human curiosity that lead to original discoveries. We are trading the chance of being extraordinary for the guarantee of being consistently average.
The survival of the “Fit” in the AI era doesn’t mean becoming the fastest prompt-user; it means becoming the person who knows what the prompt is hiding.
I have started to change how I work. I still use AI – to ignore it would be professional suicide – but I use it as a sparring partner, not a replacement. When I read a prospectus now, I no longer ask AI to “summarise the risks” first. I read the raw text, I feel the friction, I form my own “messy” opinion – and only then do I ask the AI to challenge me. I am intentionally re-inserting the rungs that the industry is trying to saw off.
Let AI agents handle the mundane, but never let them handle your curiosity. Let them process the data, but never let them define your judgement. In an age of “Zero Friction,” the heat of your own thinking is the only fire that can keep your professional self alive.
