Solo builder playbook
At ten to fifteen hours a week you cannot run a marketplace or a sales team. You pick between wiring AI into one business at a time and shipping a narrow product — and the wiki now treats both as serious, not product-versus-consulting.
The career wiki’s side-business track has two load-bearing archetypes, not one. Mark Cuban’s integrator thesis: “You do not need to build the brain. You need to build the nervous system” — walk into companies that will never hire an AI team and wire models into their operations. The niche vertical SaaS playbook: find a service people already pay $200–$500 for, deliver 70–80% of the value at 10% of the cost, win on specificity over ChatGPT. At limited hours, picking one is mostly forced — but the choice is no longer “product good, services fallback.”
The integrator path
Cuban’s market sizing, on camera: “There are 33 million companies in this country. 30 million of them are solopreneurs … millions of companies that have 1, 5, 10, 50, 100, 500 people that aren’t going to have AI budgets, aren’t going to have AI experts.” Who wires AI for them? He paraphrases Satya Nadella: “software is dead because everything’s going to be customized to your unique utilization.” The historical analogue is PCs and electricity — wealth went to people who showed factory owners where to plug in, not to the generator builders.
Arvind Jain, CEO of Glean at $200M ARR, gives vendor-side corroboration: “AI is not actually extending software in a marginal way … it’s actually going to grab a lot of revenue that actually today is in services industry which is 25 times larger than software industry.” Nick Turley, head of ChatGPT at OpenAI, names the same archetype as his highest-conviction startup bet: companies “getting extremely hands on and doing effectively professional services with AI” because labs are not proximate to domain problems. “If you get proximate, I think you can build something transformative.”
Unit economics differ from SaaS: few customers, project fees plus retainer plus rev-share, domain fluency and relationship as moat, weeks to first dollar instead of months. At 10–15 hours a week, selling one $10K–$50K SMB engagement may beat acquiring hundreds of $49/month subscribers. Cuban pitches this to new grads; whether it generalizes to principals with higher opportunity cost is unresolved on the wiki.
September 9 put numbers on that integrator row — still a draft, not a customer. A Reliquary one-pager titled “Research Second Brain — Live in 2 Weeks” prices a productized consulting install: $2,500 standard, $1,500 intro for the first two University of Florida research professors, $500 a month optional retain, $99 for a DIY kit. The beachhead is the wife’s peers. The stack is the one already running: a Karpathy-style wiki, Alfred-like ingest, two or three Grok Bot agents. Near-term goal: cover AI subscriptions; later, toward half a salary via small paid offers. Trading P&L stays with Gage. Kill rules sit on the page: zero discovery calls in fourteen days of outreach and rewrite the pitch; zero paid install by day 45 and park it or sell the kit. The clock starts when outreach starts. It has not started. Status is awaiting edits. There is no public listing and no Forge ticket. One priced SKU set does not raise the $20–40K ARR ceiling to a measured outcome, and it does not prove a solo-human company exists.
The product path
The validated micro-SaaS pattern: Photo AI replaced $200–$500 professional headshots with AI-generated shots at $29/month and reached $132K MRR in 18 months as a solo operation, per the Indie Hackers case study the wiki cites. Freemius aggregated data: median profitable micro-SaaS MRR around $4,200 (~$50K ARR); 45.7% of marketplace products solo-founded; AI-featured micro-SaaS grows roughly 2× faster early — but AI alone does not guarantee profitability (61% at breakeven vs 54% non-AI, near parity).
Honest timelines from solopreneur benchmarks: six months realistic at $1,000–$3,000 MRR; twelve months at $5,000–$15,000 MRR with profitability claims read skeptically; twenty-four to thirty-six months at $20,000–$50,000 MRR for a full-time equivalent. Horizontal AI wrappers saturate immediately; hyper-vertical products compete with paralegals and photographers, not with Anthropic. Price against the reference service, not against other AI tools. Distribution — building in public, community engagement, SEO — is the bottleneck once boilerplate auth and payments exist; paid ads are generally unsuitable at this scale.
Two archetypes at 10–15 hrs/week
Model comparison from the career wiki’s side-business question pages. Median and plateau figures from Freemius and TwoCents benchmarks.
One human, many agents
The solo-human-company thesis asks whether one operator plus AI agents can reach $500K–$1M ARR or whether ceilings stay near $100K–$200K. AI lowers the cost of creating across code, copy, and analysis; selling, negotiating, and judgment under uncertainty stay human-led on current evidence. Two routes fit the one-human shape: productize a SaaS with AI-automated support and ops — Pieter Levels runs a $1.6M ARR business alone using AI for triage and minor dev, per the wiki — or productize integration via agents and playbooks so engagements parallelize.
Failure mode to avoid: 70% of micro-SaaS products stagnate under $1,000 MRR — primarily niche selection and distribution fit, not product quality. Building commodity auth and payments from scratch wastes three to six months. The open questions the wiki keeps live: which vertical is underserved in 2026, whether integration services are transient before incumbents ship vertical AI-in-a-box, and whether the thesis survives outside Valley tooling hothouses — Andreessen’s non-coder partner anecdote is one existence proof with unlimited compute and peer support, not a base rate. Kareem Amin’s version of the same who-gets-leverage question points the programming-shaped tool at go-to-market people, not at a solo product P&L. “We started with the really abstract ambition of how do we give the power of programming to more people.” Clay is not an existence proof that one human plus agents hits $500K. Do not treat the host’s valuation as evidence the thesis works.