Staying employable beside AI: practical career moves from operators

The Short Version

Four people who run things give the same first instruction about AI and work: start using it now, on your own real tasks. Carla Harris, a Wall Street veteran, says "if you don't want to become a fossil, you have to at least start to play around with it" [1]. Chris Koerner, who has started dozens of businesses, tells people to pick work at one of two ends, either physical work that machines will not reach soon or work that leans heavily on AI, and to avoid what he calls "the messy middle" [2]. The programmer David Heinemeier Hansson, known as DHH, says agents now write his code, that the scarce skills are vision, taste and product judgment, and that someone who missed the last year can catch up in about two weeks [3]. Pieter Levels shows what one person with simple tools, automation and a payment button can build alone [4]. Their advice differs on whether to stay inside a company or build something of your own, and all of it can be acted on this year.

Table of contents

What They Said

Carla Harris: get on the dance floor, and remember who decides

Harris spent more than three decades at Morgan Stanley, and her advice is for people working inside organizations. She starts from the observation that nobody has the playbook. In the past there was a set routine for advancing in a company. Now many companies expect employees to train themselves, and leaders do not yet know what to value [1]. "This is not the time to be dictated to when you have the opportunity to design," she says [1]. In uncertain periods, "half the people are distracted, the other half are paralyzed," which leaves a clear path for anyone who acts [1].

On AI itself she is practical. "Nobody invited it in, but they're at the party," so do not sit against the wall. Learn alongside everyone else and ask friends how they use it [1]. Her examples are ordinary. She used it to build a workout and diet plan. A friend used it to prepare questions for a new specialist doctor. Another built an agent to pick out her five most important emails, which meant checking its work at first [1]. When Harris was handed a book the day before she had to interview its author on stage, AI gave her chapter summaries and three questions per chapter in under ten minutes, in place of the six or seven hours she would have spent cramming for a thirty-minute conversation [1]. Mel Robbins adds a use of her own: getting feedback on your work when your boss is a bottleneck [1].

Harris does not wave away the objections. The technology may need regulation, it is not always right, and it produces hallucinations. "But you can't afford to wait until all of that is ironed out before you engage," she says, because a year from now you may feel unable to take part in an economy where everyone uses it. "You are not too young or too old to start playing around with it" [1].

Her second point is about what remains distinctive once everyone has the same tools. When everybody has access to the same information and the same training, what separates person A from person B is an untried idea and the courage to try it, together with relationship skills, because, as she puts it, "people are making people decisions" [1]. She suggests a script for proposing something new: I am confident about this, give me thirty days or three months, and if I am wrong you can move me back. Companies, she says, "are going to value people who have clarity, people who are willing to take risk, people who can think out of the box now" [1]. She also urges people to propose new roles inside their own organizations, spelling out the reason, the benefits and the market value, since an organization does not know how to value something new unless you tell it [1].

Her older career rules matter more in a period of reshuffling. Every decision about your pay and promotion "will be made in a room behind closed doors where you are not present," so someone has to speak for you [1]. She calls that person a sponsor, and says people who keep their heads down overinvest in their deliverables and underinvest in relationships. "Hope is not a strategy" [1]. She also offers a rule for reading job postings: "Whenever you're reading a job description that says required, I want you to read preferred" [1]. And she notes from nearly forty years of watching that companies hire selectively even during layoffs, using hard economies to upgrade their talent [1].

For anyone who has been laid off or is between things, she calls the period the "meantime valley" and treats the time as a gift, even when money is tight. Identify five companies you would like to work for and why, and take a job you will be able to tell a story about in two years [1]. And she prescribes rest before effort, having ignored sleep, by her own account, for the first twenty or thirty years of her career [1].

Chris Koerner: pick an end of the barbell

Koerner's framework is about choosing what kind of work to be in. He pictures a barbell. At one end is work with "no AI, no tech, no internet, sweaty, hands on, difficult": plumbing, lawn mowing, tree removal. He likes those businesses because people do not want to do them, and you can charge a premium. "AI isn't about to disrupt that," he says [2]. At the other end are businesses that are heavy on AI and run online. He avoids "the messy middle," which can be disrupted from either side [2].

His reasoning about the physical end is specific. The people building robots, he says, are "making robots that can you know package orders and do things at scale not replace a toilet" [2]. He says America is short about three million plumbers, heating and cooling technicians and electricians, and that beyond those three there are thousands of home-service niches, down to separate businesses for gutter repair, gutter replacement and gutter cleaning [2].

He backs it with people he has met. A young couple washed his windows for $500, and he points out they could have asked for a deposit and bought their equipment with it [2]. A man who mounts televisions told him he does eight to twelve a day and earns around $300,000 a year, with no advertising, purely from profiles on free lead platforms, by doing good work and asking for five-star reviews [2]. Koerner's comment is that someone who spends a week or two learning such a skill and masters it has something "people are always going to need" [2].

At the AI end, his top opportunity for an ordinary person is consulting for small businesses. He cites surveys showing that about 85 percent of small business owners know they need AI and only about 12 percent use it meaningfully, mostly for email [2]. Roughly one adult in ten owns a business, so a person with 400 online connections probably knows about forty owners [2]. His method: take one to coffee, record the conversation, ask what is frustrating about the business and keep asking why. Upload the transcript to an AI system, ask for suggestions, choose one, and have the AI walk you through carrying it out. At the next meeting you are the most AI-capable person that owner knows, and he says they will tell you to "take my money" [2].

His general rules apply to both ends. "People are masters at playing business and not doing business," he says, meaning they form a company and design a logo when they should be finding a customer [2]. He prizes a bias for action above other traits, and trains it by investigating every question immediately, for instance by asking an AI model on a walk why some railway sleepers are wood and others concrete [2]. He tells his children that "no one's coming to save you," and treats the ability to earn independently as the family's equivalent of stored food [2].

DHH: from writing code to directing it

DHH created the Ruby on Rails framework and has written code for 25 years. A year before this conversation he was a skeptic of AI coding. Now he says all the new code in the latest release of his Linux distribution was written by agents he steered [3].

He dates the change to late 2025, when a new model's output came uncannily close to what he would have written himself, and he credits better tooling as much as better intelligence [3]. He describes phases: first instructing agents step by step, then agents that split a task among sub-agents and cut the time to a fifth or a tenth, then models that choose the route themselves from a loose description [3]. "I've become optional in the part that produces the code that picks the route," he says [3]. Asked whether he is suffering from a kind of AI mania, he answers, "Of course I'm delirious," and points to the fact that he ships software people download [3].

From that experience he draws advice for anyone anxious about a technical career.

Know which part of the job you loved. If the only thing you loved about programming was the mechanical assembly of logic to build what others told you to build, he says, you will struggle, "because that mechanical process is under threat." But "if you are as you just mentioned excited about building things i don't think you're under threat at all." He thinks there is a strong case that far more builders are needed [3].

Direct, and stay loose. Product-management skills now matter more than typing skills. He advises being as vague as possible at the start and discovering what you want by using what comes out, and warns that deep programming knowledge can become a handicap if it leads you to prescribe a path the agent would handle better [3].

Cross-check. His routine is to drive with one model, review with another, and let a third catch more problems. He cites a study at Shopify that traced production incidents back to the changes that caused them and found that changes reviewed by agents caused fewer problems than those reviewed by humans [3].

Work with the agents yourself. Asked why large companies have not visibly sped up, he says writing code was rarely the bottleneck. Human bandwidth, layers of approval and a shortage of vision and taste are [3]. To get the large gains you have to work with agents directly and not through other people. He has reached the point where "the human in the loop is the limit here" [3].

Build the small thing you need. Big companies are slow to turn, he says, which leaves room for individuals to rebuild the small fraction of a large program they actually use. He replaced a writing app with one built by agents in about twenty minutes [3]. When Lex Fridman says he finds it intimidating to release a tool he built for himself, DHH says it is simpler than he thinks: tell the agent to publish it, and it will write the documentation too [3].

Do not try to predict. Even the smartest people cannot see two model generations ahead, so focus on the present and lean in [3]. Someone who missed the last year can reach the frontier in about two weeks [3]. Community helps against dread, and it is acceptable to grieve the old craft for a while [3]. "You have a choice to a large degree of how to lean," he says [3].

He is not blind to the cost. He cites the economic argument that cheaper programs mean more demand for them, and the example of cash machines, which were followed by more bank tellers. He also says productivity gains do displace people, and that it is tragic for the individual who is laid off [3]. Fridman presses the darker possibility that there may eventually be no programmers at all, and stresses compassion for those who suffer in the transition [3].

Pieter Levels: what one person can run

Levels has launched more than forty small internet businesses, mostly alone, and his example bears on the question of what a single worker can now do. "I don't really raise money. I don't use VC funding. I do everything myself," he says [4]. He builds a small product, launches in about two weeks with a payment button from the first day, and finds out whether people will pay. Most ideas fail, and he expects that [4].

His AI businesses began with play. When an image-generating model appeared, he noticed it was good at houses, made a toy site, then turned it into an interior design tool that earned $10,000 to $20,000 a month within a week [4]. A tool that made portraits from a user's own photographs earned about $150,000 in a week before larger companies copied it [4]. He launched that with one web page, a payment link and an upload form, and trained the first few hundred models by hand until the workload forced him to automate [4].

Automation is how he avoids hiring. His sites run on scheduled jobs. Status pages show green or red, a monitoring service alerts him on his phone within a minute, and every software error is sent to him the same way [4]. He uses a language model to fight spam and moderate a chat of ten thousand people, and says it understands humour and double meanings better than human moderators did [4].

He is unbothered by fashion in tools. He uses a plain, old technology stack because it "just stays the same and works," and says tools matter less than knowing them well [4]. He charges from the start and reckons a thousand paying customers at $30 each is enough for a solo maker to live well [4].

His career began from a low point that resembles what many fear now. At 27 he was earning little, alone in hotel rooms, and depressed. His father's advice was physical: get a shovel and move a pile of sand. "You can't just sit still." He set himself the challenge of launching twelve small businesses in twelve months, which taught him to code [4].

Where They Agree, and Where They Don't

Use it now. Harris, DHH and Koerner all say the cost of waiting is higher than the cost of imperfect tools [1][2][3].

Action beats planning. Harris's paralyzed half, Koerner's people playing business, DHH's advice to iterate and Levels's two-week launches make the same point [1][2][3][4].

Human judgment and relationships keep their value. Harris names relationships and courage, DHH names taste and vision, and Koerner's businesses run on reviews and personal networks [1][2][3].

Small can now compete with large. DHH, Levels and Koerner each describe one person doing what used to need a team [2][3][4].

They differ on four points.

Inside a company or outside it? Harris coaches people to advance within organizations through sponsors and new roles [1]. Koerner and Levels advise building your own income, and DHH thinks large firms are too slow to capture the gains [2][3][4]. The advice reflects where each made a career. Harris also notes that companies keep hiring through downturns [1].

Can the middle be saved? Koerner would leave it [2]. Harris's listeners mostly work in it, and her answer is to become the person with relationships, ideas and the nerve to propose them [1]. DHH's version is to move up from producing to directing [3].

Predict or not? Koerner's barbell is a forecast about which work is safe [2]. DHH says nobody can forecast two model generations out and that you should not try [3]. Koerner's bet on trades depends on robots continuing to lag, and Fridman wonders aloud whether service work may be what is left [2][3].

Newest tools or familiar ones? DHH runs many agents in parallel and ranks the latest models [3]. Levels has used the same simple stack for years [4]. Both ship constantly, which suggests the habit of shipping matters more than the choice.

A further difference is in tone about loss. DHH and Harris are exuberant. Fridman insists on the pain of those displaced, and DHH agrees it is real [3].

Why It Works

Shared tools shift value to what is not shared. When everyone can get the same output, the differences are ideas, judgment and trust [1][3].

Implementation got cheap. If producing the code or the draft is no longer the constraint, the person who knows what should be built and can check it becomes the constraint [3].

Physical work in varied settings is hard to automate. That is Koerner's reason for the hands-on end [2].

There is a gap between knowing and using. Most small firms know they need AI and few use it well, which is a market for anyone slightly ahead [2].

Automation substitutes for staff. Levels's monitoring and moderation let one person run several businesses [4].

Hesitation by others clears the field. Harris's observation is that most people freeze in uncertainty [1].

The evidence is the operators' own results and a few cited figures: Koerner's survey numbers and labour shortage, DHH's output and the Shopify finding, Levels's revenues, and Harris's decades of observation [1][2][3][4]. These conversations took place at different times, and Levels was describing an earlier generation of tools than DHH [3][4].

Where It Applies, and Where It Doesn't

DHH's account is about software. His claims about agents come from programming, where output can be run and tested [3].

Harris's advice assumes a workplace with sponsors and reviews. It fits corporate careers best [1].

Koerner's examples are from American home services. The shortage figure and the lead platforms are specific to that market, and the work is physically hard, as he says himself [2].

Solo building is not for everyone. Levels describes loneliness and depression along the way, and most of his projects failed [4].

AI output needs checking. Harris's friend had to verify her email agent, and DHH reviews with second and third models [1][3].

Nobody here promises safety. DHH says displacement is real and the future cannot be predicted [3].

What To Do With It

1. Use an AI tool this week on a real task. Pick something that costs you hours, such as preparing for a meeting, summarizing a long document or drafting questions, and check the result [1].

2. Ask three people how they use it. Harris's method for learning without a playbook is to copy what works for friends [1].

3. Decide which end of the barbell you are moving toward. Either learn a hands-on skill that people will always need, or make your work AI-heavy, and plan your way out of the middle [2].

4. If you write code or produce documents, practise directing. Give a loose brief, look at what comes back, refine, and have a second model review the first [3].

5. Give it two weeks. If you feel behind, set aside a fortnight of daily use, on DHH's estimate that this is enough to reach the frontier [3].

6. Sell what you learn to someone who has not learned it. Try Koerner's coffee meeting with one business owner you know [2].

7. Build one small tool for yourself and publish it. Replace the part of a big program you actually use, or automate a chore, and charge from the first day if others want it [3][4].

8. Automate your own monitoring and routine. Set up alerts and scheduled jobs so that you learn of problems without watching for them [4].

9. Find a sponsor and propose something. Identify who speaks in the rooms where your pay is decided, and bring an idea with a time limit attached [1].

10. Rest, then move. Harris prescribes sleep before effort, DHH prescribes company and permission to grieve, and Levels's father prescribed a shovel [1][3][4].

Sources

1. Carla Harris on the New Rules of Work: Sponsors, Confidence, Raises and AI in an Uncertain Economy · The Mel Robbins Podcast
2. Chris Koerner on Starting a Business With No Money, the Barbell Method, and Donating a Kidney to a Stranger · The School of Greatness
3. DHH on Agentic Engineering, Omarchy Linux, and Why He Stopped Hand-Writing Code · Lex Fridman Podcast
4. Pieter Levels on Building Solo Startups, Photo AI, Simple Code and Digital Nomad Life · Lex Fridman Podcast