AI Can Build Your App. It Can't Make It Good. | Paul Bakaus, Impeccable

Solo Founders 1h21 4 min #26
AI Can Build Your App. It Can't Make It Good. | Paul Bakaus, Impeccable
Watch on YouTube

Summary

  • Paul Bakaus, founder of Renaissance Geek and creator of Impeccable (an AI coding-agent skill for design), shares his journey from a teenager in Germany’s Black Forest to building jQuery UI, a game-engine startup acquired by Zynga, and roles at Google — then explains why he chose solo founding with AI, how Impeccable keeps AI-generated design on-brand, and why human judgment (editing, subtraction, taste) remains the irreducible bottleneck in an era of abundant generation.

Background and early path to the open web

  • Grew up in the Black Forest, Germany, far from tech hubs; the early web was a “normalizing factor” that let him reach the frontier from his basement.
  • Started a consulting business at age 11–12, requiring a special hearing at the city office with his mother present.
  • Discovered jQuery as a teenager; it lowered the gate for non-programmers to build interactive experiences, and he became a core contributor.
  • Created jQuery UI to solve the lack of rich UI components (drag-and-drop, accordions, etc.), giving him a front-row seat to the Web 1.0 → 2.0 transition.

Game-engine startup and Zynga acqui-hire

  • Co-founded a game-engine startup (~2010) with a former agency colleague to let every web developer become a game developer; built an isometric engine that ran on the original iPad.
  • Distribution and monetization were unsolved; European investors (Samwer brothers) only backed proven copycat models, making fundraising nearly impossible despite strong inbound interest from Microsoft, Zynga, and others.
  • Ended in an acqui-hire by Zynga; spent two years in R&D, learning virality mechanics (K-factor, whale economics, forced social loops) at massive scale.

Why solo founding this time

  • After climbing to upper-middle management at Google and other roles, realized he missed building; didn’t want to become a middle manager again.
  • Wanted to test the “solo unicorn” hypothesis: can one person 100x their output with AI agents?
  • Built extensive scaffolding tooling and took on contract projects meant for 5–10-person agencies to prove the model; confirmed it works, but attention is the new bottleneck.
  • Peter Steinberger’s “attention is the only thing you can’t scale” insight resonated: optimize for bursty, high-leverage creative work.

Hiring a chief of staff, not engineers

  • First hire is a chief of staff / operations hybrid — deliberately not an engineer, designer, or PM.
  • Rationale: if you’re the best builder, giving that up to manage people is a trap; hire someone to remove operational clutter so you stay on the tools.
  • Job description emphasizes low ego, willingness to do grunt work, and ability to scale from “set up bank accounts” to “negotiate strategic partnerships.”
  • Warns that big-company operations specialists often fail in early-stage chaos because they’ve only owned one narrow slice.

What Impeccable is and how it works

  • Impeccable is a “skill” (extension) for coding agents (Claude Code, Codex, GitHub Copilot) that makes them better at design.
  • Two goals: raise the floor (decent baseline design for non-designers) and raise the ceiling (help experts go from good to great).
  • On init, runs a product interview (audience, intent, context — e.g., kids vs. parents for a Lego app) and generates/updates a design.md (tokens, typography, components) based on Google’s Design MD spec.
  • Uses design hooks (linting) that run after the agent writes code: detects off-brand colors, fonts, AI slop, and forces the agent to self-correct transparently.
  • Every rule is ablation-tested in an eval harness; rules that don’t reliably improve output are removed.

From side project to a16z-backed company

  • Two parallel signals: (1) conversations with partner Anish at a16z convinced him his idea arsenal had enough surface area for something to work; (2) organic “why is this free? I want to pay” tweets on Impeccable — a rare, strong PMF signal.
  • Released multiple OSS projects earlier in the year; Impeccable was the one that caught fire.

Free-first growth strategy

  • After years of neglecting Twitter, found the algorithm ignored his new AI/design content because his follower base was from jQuery/creator-economy eras.
  • Strategy: create maximum value, give it away free, optimize for momentum and organic K-factor (what users say when you’re not in the room).
  • No paid ads, no influencer campaigns; FireShip video and other coverage were fully organic.
  • Believes forced virality (Zynga/Facebook style) is unpleasant and misaligned with a craft-focused business; prefers a smaller, loyal core over a leaky funnel of drive-by users.

Integrity, quality, and “be kind, not nice”

  • Personal integrity > growth hacks: would feel like a “snake oil salesman” if someone used his tool based on a claim that didn’t hold.
  • Most AI skills are “vibe-coded” — tested once or twice on the author’s machine, then shipped; Impeccable is rigorously eval-tested.
  • “Be kind, not nice”: give real, sometimes uncomfortable feedback rather than empty pleasantries; respects the recipient’s growth.

Where humans belong in the loop

  • Three phases where human judgment is critical:
    1. What phase (product): defining the problem, audience, intent — if you don’t know why you’re building, why delegate to an agent?
    2. Building phase: agents drift randomly over long horizons; periodic check-ins prevent catastrophic divergence.
    3. Polishing phase (last 10–20%): subtraction, simplification, taste — “the most important skill right now is to remove things.”
  • Steve Jobs wasn’t a “creative genius” in the generative sense; he was the greatest editor — said no to 99%, shipped the 1%.
  • Analogy: working with LLMs is carving software out of marble (or navigating latent space) — the model contains everything; your job is to reveal the right shape by cutting away.

Artisan vs. factory defaults

  • “Pretty good” is a moving target (Bootstrap → Squarespace → Claude defaults); in a sea of vibe-coded sameness, standing out requires intentional, opinionated design.
  • Context matters: enterprise tools should feel familiar (system fonts, copied patterns); consumer landing pages have 5 seconds to differentiate.
  • Impeccable’s visual iteration mode lets humans steer variants rapidly — keeping the human in the loop for the judgment calls AI can’t make.

Bear case for solo founding

  • “It takes a certain type of crazy to found something… not knowing whether you’re actually crazy or not” — lack of a sounding board is the biggest risk.
  • Mitigate: find objective peers, not just friends/family.
  • Protecting attention is harder alone; contracting helps but doesn’t fully solve it.

Case for solo founding

  • You only account to yourself; move at the speed of your own thought.
  • Don’t wait for the “right co-founder” to appear — start now; the right people (chief of staff, collaborators) come later.
  • Momentum cures burnout: focus daily on “what value did I create today?” and hire only people who create daily value without middle management.

Closing thought

  • The era of the “renaissance geek” (40-year-old founder with cross-functional intuition + AI leverage) is here: less energy than a 20-year-old, but far fewer unknown unknowns, and the tools to shape whatever you want directly.
Back to Solo Founders