the economy split

here's the $40 trillion secret nobody talks about.

while everyone's panicking about ai taking jobs, solopreneurs and micro-businesses quietly generate half of global gdp. that's more than china's entire economy.

while junior developers can't find work, solo founders build products that would've needed $2m seed rounds five years ago. for $150/month.

while consulting firms collapse and law firms stop hiring juniors, millions of people are finding weird little niches that didn't exist last year.

something doesn't add up.

or maybe it does.

sorry for the radio silence recently, mostly spent time building, reading and reflecting...

actually just building… and finally got 10mins to reflect on what i’ve been building and reading. here are some of my latest finds/thoughts on ai, the (job) market and what i’ve been building.

no time to read? check out this video that explains it all

so what’s up with jobs and the job market?

there are two schools of thought out there:

one is that we're f&%#d

and the other… that history has proven that this is just another transition period. one leading to greener pastures.

twitter thinks ai will either create infinite jobs or destroy them all.

after running ten ai transformations for scaleups or major corporations, i can tell you what's actually happening behind closed doors. every boardroom conversation ends the same way: "how do we do more with less people?"

they're not being evil. they're being rational. when one developer or marketer with ai can do what five did last year, why would you hire five?

i know because i'm playing both sides. mornings, i help corporations automate. afternoons, i build ventures with the same tools that are eliminating their headcount. what used to need a $2m seed round and a team of 20 now runs on $150/month and pure focus.

the hypocrisy? maybe. but it's given me a view nobody else has.

here's the pattern i see everywhere: companies aren't replacing workers with ai. they're just not replacing workers. period. attrition becomes strategy. the junior role that opens when someone quits? it doesn't get posted. it gets automated.

meanwhile, the real money isn't even in ai. nvidia's market cap jumped $2 trillion. aws prints money selling compute. the ai companies themselves? most are burning cash building features that become free six months later.

for every 20usd/month we give to cursor, half goes to openai/anthropic, another 5usd goes to nvidia and only the remaining stays with cursor.

everyone’s building wrappers within wrappers within wrappers :)

here’s what the supply chain looks like

btw follow me on x/twitter, i’ve finally started sharing insights regularly: https://x.com/darnocks

look at what happened since chatgpt launched in november 2022. openai couldn't handle demand. they had to use supercomputer clusters meant for training new models just to keep the service running. nvidia added $2 trillion market cap because there literally weren't enough computers on the planet.

this isn't like when factories replaced farms. or when excel replaced ledgers. those transitions moved people from physical work to cognitive work.

if you want to understand where this is headed = not the twitter version, but what's actually happening in the rooms where decisions get made… i can show you.

because the truth is weirder than both the doomers and optimists think.

ai is coming for cognition itself.

ok sure…the optimists have compelling data. in the previous tech revolution, the u.s. gdp doubled from $11 trillion to $27 trillion in 20 years despite constant disruption.
- amazon employs more people than gm.
- apple has more employees than mcdonald's.
- alphabet has 182,502 employees, gm has 163,000. tech didn't destroy jobs, it created different ones.
- at microsoft, less than half the employees are software engineers.
- apple bet on human staff in physical stores when everyone said retail was dead. they now have greater net profit than meta and google combined at $100 billion.

humans plus tech beat pure tech.

true? maybe..

but i believe this time is different.
not because ai is magic.

but because it's the first technology that directly competes with human thought.

what do the numbers currently say? if you own equity or started a company pre-ai boom, you're winning.

if you’re in the workforce… things don’t look great

entering the workforce now, you're struggling.

what about developers?

senior engineers stay valuable because they understand systems. juniors face harder competition because ai codes better than them.

junior developers can't find jobs because ai writes their code better.

here are a few more sources:
entry-level tech hiring is down 50% vs pre-covid (signalfire, 2025). 

overall developer postings sit at 65% of feb 2020 levels (indeed/fred, 2025). fewer openings, more competition, and ai eats the work juniors used to cut their teeth on.

consultants?

check out this video where i explain that the consulting industry is seriously contracting, and things will only get worse.

the message? consulting work is real, but quieter and junior roles are getting squeezed. which essentially makes the whole food chain contracts

what about lawyers?

law school?

tarifi might be right. goldman sachs predicts that up to 44% of legal work tasks could soon be automated

junior associates are taking the biggest hit. makes sense right? document review, contract analysis, and legal research are vanishing into ai's black box.

brad karp, chairman of paul weiss, recently predicted that junior lawyers will be "supplemented, if not significantly replaced" by ai, technologists, and data scientists

long story short… ”the composition of law firms is evolving, with a shift towards more experienced lateral hires, growth in two-tier partner structures, and a reduction in junior associate hiring,”

and the list goes on…

so what to do?

protect jobs at all cost? i read last week about the the ‘license raj’ (a system of heavy government regulation and control) that stalled india's development for decades after 1948 by overprotecting jobs. europe's tech-skepticism contributed to growth rates half of america's. but america in 1838 had 80% of people farming. now it's 1%. those farmers' descendants didn't become unemployed. they became everything else.

nobody in 1838 could imagine james cameron working with 4000 people on avengers endgame, a film costing more than the entire 1838 u.s. military budget. the number of people who worked on that one movie was more than half the size of the entire u.s. army back then.

so yes, the future is bigger than we can imagine.

maybe ai creates massive inequality. maybe it creates massive abundance. but what usually hurts is the transition period. that period in between.

the transition

the ‘messy passage’

history reminds us: it’s the in‑between that bruises hardest. mit economist david autor finds that since the 1980s, automation has wiped out more jobs than it’s created-especially in higher-skilled roles-because machines aren’t just replacing poor jobs anymore, they’re hitting white‑collar ones too. his research, suggests we may have reached an inflection point whereby technology is already destroying more jobs than it is creating.

and deming, ong & summers show that past waves like electricity or steam didn’t destroy jobs overnight, they shifted them over decades, and the pace today is just catching up. the “messy passage” is getting messier and faster.

because with all the talk of job retraining and changing careers, issue is that results of these programs aren’t great

what about job retraining?

the data on job retraining is mixed. the u.s. spends $20 billion annually on job programs with weak results. the eu spends well over $20billion (equivalent) annually on job retraining, with moderate but real success…especially for long-term unemployed and when retraining is tied directly to ongoing employment or apprenticeship. retraining is less likely to work if detached from current jobs or when entire industries disappear at once (which therefore matches u.s. patterns. so you might be thinking “but people switch careers successfully all the time”. sure but retraining works when you're still employed and learning on the job. it fails when entire towns lose their only factory.

where’s my head at in all this?

i do believe we're watching a k-shaped recovery become a k-shaped economy. the split between capital and labor is real.

but here's what excites me: if capital is eating labor, then the logical move is to become capital.

so…

not in 5 years when you're "ready." not after another certification. now.

why? because the tools that used to require teams of 50 now need teams of 5. or 1. the same ai that's eliminating junior developer jobs is also eliminating the need for junior developers in your startup.

i've watched founders build products solo that would've needed $2m seed rounds just 5 years ago. they're doing it for $150/month. cursor, claude, v0… the entire stack costs less than a gym membership.

this realization pushed me into the venture studio phase of my career. not because i wanted to build one big thing, but because i can now build many things. fast. the friction that kept ideas as ideas is gone. deep down i always wanted this, but the idea of building a large dev team felt too cumbersome. that time has changed. you can now build with a minimally-sized team (or alone).

don’t believe that with the right stack you can build digital ventures for under 150usd/month? i share our entire stack in this video

enter humanoidz

this opportunity truly kickstarted the launch of humanoidz (which i mentioned in a previous post).

what do we do? we launch ventures. (follow our journey step by step by joining our discord here)

how?

1 = we partner with ‘secret-holders’ = corporates, holding companies or family offices that are sitting on insights but lacking execution. they have the problems, we have the building capacity. we build ventures 50/50.

2 = we build our own internal ventures. tools for the solopreneur explosion. because if everyone needs to become capital, everyone needs the tools to discover, validate, build and scale.

i’m convinced we’ve entered the builder’s era. an era where markets will become increasingly scattered = a few giants + millions of smaller players (even more so than today)

the real opportunity

watch what's happening to incumbent saas. salesforce, monday . com » their market caps are starting to crack. why? because when anyone can build software, why pay $50k/year for a crm when you can build exactly what you need for your specific use case?

the market isn't consolidating. it's atomizing.

which means the market will get much more fragmented, leaving place for a much larger number of smaller businesses focused on niche applications. millions of micro-saas products for millions of micro-niches. the long tail of software is about to get very, very long.

here's data we rarely talk about: solopreneurs and micro-businesses (under 10 employees) generate more than $40 trillion per year globally, that's about half of global gdp, that's larger than the gdp of china... but schools still train people for corporate jobs.

a year ago i mapped out what an ai-augmented builder stack looks like. back then it felt futuristic. now it feels obvious. the tools we thought were revolutionary are already being commoditized. which means the real value isn't in the tools = it's in knowing what to build with them.

here is an example of what an ai-augmented stack could look like. i made this about a year ago.

and here’s a rough version of our humanoidz roadmap (it’s changed a bit but i haven’t taken the time to update it)…you’ll have to click to zoom in. we’re also sharing the steps of our journey in this discord which you’re invited to

step 1 = we first built www.dontbuildthis.com… to start challenging our own roadmap and now we’ve made it public to challenge what people are building whilst also offering better alternatives. it’s become a treasure trove of ideas and a community of builders.

here’s our july report for what people are building

we then built www.builderbox.ai … because vibe-coders are hitting code quality walls after a couple months.

check out the video here:

we’re also building a “nuggetfinder” = a business idea generator (coming soon) that tackles the real bottleneck: most people don't lack building ability, they lack problem-finding ability. they can code but can't spot opportunities.

again if you’re like to follow our journey at humanoidz join our discord

and we’ll continue to develops solutions/apps along the builder journey… discover, validate, build, market, scale, optimize.

coming back to the uncomfortable truth

we're not going back to a world where labor commands premium value. that ship has sailed. but we're entering a world where a single person with taste, domain knowledge, and the right stack can compete with entire companies.

the question isn't whether ai will take your job. it's whether you'll use ai to build something that makes jobs irrelevant.

so what's your move?

if you have capital, compound it aggressively. every dollar invested in building capacity today is worth ten dollars waiting for the "perfect moment." the asymmetry has never been this extreme.

if you have time and energy, start building. not tomorrow. today. even if it's small. even if it sucks. the first version always does. but the cost of trying has collapsed while the cost of waiting has exploded. every month you wait, another thousand builders enter the market with tools that get 10% better and 50% cheaper.

if you're still in the traditional workforce, treat it like a paid education. learn everything. document everything. steal processes, understand problems, spot inefficiencies. your job is your market research lab. use it before it uses you up.

speaking of learning. the speed at which you can acquire new skills now determines your survival velocity. not your degree. not your years of experience. your learning rate. i've been obsessed with accelerating this. tried every course, bootcamp, youtube rabbit hole. most are built for a world where you had months to learn something. we don't have months anymore. we have weeks. sometimes days. so i engineered this prompt that turns any ai into a feynman-inspired learning coach. it's what i use when i need to go from zero to dangerous on any topic, fast. steal it:

Act like Richard Feynman–inspired master teacher, cognitive-science learning coach, curriculum designer, and accountability partner for a single learner. You make complex ideas obvious using everyday analogies, thought experiments, and plain language, while building a realistic study plan and schedule that adapts to the learner.

OBJECTIVE You will teach ANY subject the user names from first principles to practical competence, then to deeper expertise. You will: • map the core fundamentals (the “vital few”), • design a personalized learning path, • create a concrete study schedule, • deliver lessons with analogies and checks for understanding, • assign projects and retrieval practice, • adapt as the learner progresses.

FIRST ACTION (before planning) If the user has not provided the following inputs, ask them as a numbered list and wait for answers before proceeding:

1. Subject or scope (e.g., “linear algebra for ML” vs. “full linear algebra”).

2. Desired outcome and use-case (what they need to do with it).

3. Current level (novice / beginner / intermediate / advanced) and any prior knowledge.

4. Time budget (hours/day and days/week) and total timeframe or deadline.

5. Learning preferences (video/text/coding/projects), constraints (budget, device, offline), and accessibility needs.

6. Motivation and stakes (exam, job, hobby), tolerance for math/notation, and preferred tone/pace.

7. Tooling available (software, data, lab access) and languages (instruction + resources).

8. Assessment preference (quizzes, projects, oral explanation). If any item is omitted, propose sensible defaults but clearly label them as assumptions and confirm.

OUTPUT FORMAT & STYLE REQUIREMENTS • Use clear section headers, short paragraphs, and concise bullet points. • Define every new term in one plain sentence; then add an everyday analogy. • For each key idea, give: (a) ELI5 explanation, (b) everyday analogy, (c) one-sentence “why it matters,” (d) a 2–5 minute micro-exercise. • Use Socratic prompts (“What would happen if…?”) and encourage teach-back in simple language (Feynman technique). • Include periodic “Common pitfalls & how to avoid them” and “Contrast with similar ideas.” • When giving steps, number them. When listing resources, annotate why each is chosen. • If browsing is available, fetch up-to-date resources and cite them; if not, provide evergreen options and mark anything that might be outdated. • Default to friendly, curious, and rigorous tone. Avoid jargon; when necessary, translate it.

WORKFLOW (follow in order) Step 1 - Calibrate success • Restate the user’s goal in your own words and propose 2–3 measurable success criteria (e.g., “solve 20 representative problems in 45 minutes with ≥90% accuracy” or “ship a demo app using X in two weeks”). • Identify prerequisite knowledge and how you’ll cover gaps.

Step 2 - Rapid diagnostic • Give a 5–8 question baseline check (mixed: multiple choice + 1–2 short “explain in your own words”). • Score quickly; map the learner to a level (A0–C2 or Novice→Expert). Summarize strengths/weaknesses.

Step 3 - 80/20 fundamentals map • List 8–15 “vital few” concepts/skills that unlock 80% of results. • For each item: name → plain definition → Feynman-style analogy → one tiny exercise → 1 common pitfall.

Step 4 - Learning roadmap (milestones) • Break the journey into 4–7 milestones (M1…M7). • For each milestone: outcomes, what to be able to explain/do, dependencies, quick self-test (3 questions), and a mini-project suggestion.

Step 5 - Personalized study schedule • Compute total hours from the user’s time budget and timeframe; include 1 rest day/week. • Build a week-by-week plan (W1…Wn) and a sample day plan (e.g., 45–60 min blocks). • Allocate time by ratio: Learn (40%), Practice (40%), Review/Reflect (20%). Adjust based on diagnostics. • Include spaced repetition and retrieval practice cadence: review at +1, +3, +7, +14, +30 days. • Add “maintenance mode” plan after the main timeline. • Present schedule in a clean table with dates (if the user provided a start date).

Step 6 - Lesson template (repeatable) For each lesson in the plan, include:

1. ELI5: one-paragraph plain explanation.

2. Analogy bank: 2–3 distinct everyday analogies.

3. Worked example (step-by-step).

4. Socratic prompts (3–5 questions) to surface misconceptions.

5. Quick check (3 questions with answers).

6. Micro-exercise (5–10 min) and one stretch challenge.

7. Teach-back task: learner explains it to a bright 12-year-old; you point out gaps and simplify wording.

Step 7 - Projects & portfolio • Provide 3 tiers of projects (beginner, intermediate, capstone) aligned to milestones. • For each: brief spec, success criteria/rubric, suggested timeline, and a “make it real” extension. • Encourage publishing or demoing (GitHub, blog post, short video). Add peer-review checklist.

Step 8 - Resources (curated & annotated) • 3–5 core resources (book/course/playlist/doc) in a recommended order; annotate what each is best for. • 3 “reference sheets” or cheat sheets. • 1–2 communities (forums, Discord, StackExchange tags) for help. • If the subject is tool/library/version-sensitive, check recency and note versions; warn about outdated content.

Step 9 - Assessment & feedback loops • Weekly quiz or short oral check (you ask, learner explains). • Biweekly mini-retro: what clicked, what confused, what to change. • Dynamic difficulty adjustment: if accuracy >85% and time-on-task < planned, increase challenge; if <70%, slow down and insert remediation.

Step 10 - Troubleshooting guide • Plateaus: vary task difficulty, switch modality, interleave topics, or do a mini “speed-run” review. • Overwhelm: narrow scope, reduce inputs, extend timeline, or add “minimum viable path.” • Motivation dip: connect to user’s goals, add small wins, or pair with a public commitment.

Step 11 - Long-term retention • At the end, output a 30–60–90 day review plan, an Anki-ready deck outline (Q/A format), and a “one-page mental model” summary.

SUBJECT-SPECIFIC ADAPTATIONS (pick what fits) • Math/theory-heavy: emphasize intuitive visuals/diagrams (ASCII if needed), proof sketches, and unit-checking; include problem sets with incremental difficulty. • Programming/CS: alternate concept → code-along → small refactor → tiny test; include debugging drills and reading others’ code. • Data/ML: insist on datasets, evaluation metrics, error analysis; include a baseline→improved model iteration. • Languages: Comprehensible-input blocks, high-frequency vocab sets, substitution drills, and speaking prompts. • Business/PM: caselets, stakeholder maps, KPI trees, and simulation role-plays. • Creative skills: imitation→variation→original creation loop; critique checklists.

DELIVERABLES (after you receive the initial answers) Produce these sections in order: A) Personalized Overview (goal, success criteria, assumptions to confirm). B) Diagnostic Results (level, gaps). C) 80/20 Fundamentals Map. D) Roadmap & Milestones. E) Study Schedule (calendar table and daily template). F) Lesson 1 (full template) + preview of next 2 lessons. G) Projects & Rubrics (3 tiers). H) Resources Pack (annotated). I) Retrieval & Review Plan (spaced schedule + sample flashcards). J) Next Checkpoint (when/how we’ll measure progress). K) One concise “Explain it back to me” prompt to start the Feynman teach-back.

INTERACTION RULES • Ask only the minimum necessary questions first; then pause. • In each subsequent message, end with exactly one clear next-step question. • If the learner reports confusion, switch to a simpler analogy and create a 3-question micro-check before proceeding. • If the learner is ahead of schedule, accelerate: combine lessons, add a harder project, or deepen theory. • If any assumption changes (time, deadline, level), regenerate the schedule and highlight differences.

QUALITY BAR • Be specific and exhaustive. Aim for a thorough first plan (approx. 1,500–2,500 words) while keeping sections scannable. • Do not hallucinate facts; where uncertainty exists, say so and propose how to verify. • Keep the voice approachable, curious, and precise-like Feynman at a chalkboard.

KICKOFF (what you say now if inputs are missing) “Great-let’s tailor this. Reply with:

1. Subject/scope,

2. Desired outcome,

3. Current level,

4. Time/day & days/week,

5. Total timeframe or deadline,

6. Preferences/constraints,

7. Tools available. I’ll run a quick diagnostic and build your roadmap and schedule.”

Take a deep breath and work on this problem step-by-step.

let's conclude

the pessimists might be right about disruption. the optimists might be right about growth. but here's what i've learned after 10 ai transformations and countless conversations with founders, employees, and everyone in between:

we're not heading toward mass unemployment. we're heading toward mass reinvention.

the real pattern

every founder i work with started scared. every employee who successfully transitioned began confused. every solopreneur who's now thriving first felt overwhelmed.

the difference wasn't talent. it was movement. small, imperfect, consistent movement.

make progress…

maybe you build something. maybe you learn something. maybe you just observe the patterns and position yourself accordingly. maybe you find ways to augment what you already do well. maybe you become the translator between ai and humans in your field. maybe you're the one who figures out what ai can't do and doubles down there.

remember those $40 trillion generated by micro-businesses? that's not venture capital. that's millions of people finding their own weird little niches. the massage therapist who built a booking app. the accountant who automated her practice and now helps others do the same. the teacher who created personalized learning tools.

they're not all trying to be the next unicorn. they're just solving problems they understand deeply. with tools that cost less than their netflix subscription.

yes, the transition is messy. yes, entire career paths are evaporating. but humans are absurdly adaptable. we survived ice ages, plagues, world wars and the invention of tiktok (to be confirmed).

we'll survive this too.

not because we're special. but because we've always been good at finding the cracks where value lives. the spaces machines can't reach. the problems only humans notice.

how to position ourselves?

maybe it's building. maybe it's learning. maybe it's waiting and watching. maybe it's all three.

but whatever you do, don't freeze. the only truly wrong move is standing still while the ground shifts beneath you.

and honestly? that's always been true. we just couldn't see it as clearly before.

welcome to the permanent beta.

david

⬛️