Il a lancé un business sans rien faire (ou presque) - Pierre-Louis Biojout

Business-focused summary (Nanocorp + agentic strategy)

Core premise: “Agent-native” startups

Pierre-Louis argues that company building is shifting from “software built for humans” to systems built for AI agents:

• Small teams can win by redesigning products and workflows from scratch to be agent-compatible.
• Companies are increasingly like “codebases”—not just apps, but the surrounding operational context (documents, tools, processes) that agents can access.

Nanocorp: what the product does

Nanocorp is positioned as a platform that lets users launch and run businesses autonomously using AI agents (rather than human labor for every step). Practically, it automates:

• Product creation (SaaS/apps/digital products/training sites)
• Operations (go-to-market assets and customer support workflows)
• Deployment + iteration (continuous code/product changes)

Positioning example

• A plumber in Sacramento uses Nanocorp to generate a “back office” website/tool handling customer support and go-to-market, while the plumber focuses on plumbing work.
• The platform can also support multi-region duplication, e.g., one operator running country-by-country operations across Europe for lead generation.

Agent framework & architecture concepts (playbook-style)

The interview explains “agentic” systems using operational terminology:

• Agent definition

  • An entity that observes the environment + acts to achieve a goal.

• Harness loop (goal-driven multi-step execution)

  • Instead of “ask → answer”, an agent runs multiple steps toward an objective using tools.

• Recursive self-improvement

  • The agent can change/optimize its own outputs via iteration (with caveats about correctness).

• Tool/connector layer

  • Agents require integrations (email, invoices, storage, accounting, etc.). Without tool access, they can’t do real work.

• “Context must be digitized”

  • Best results come when business knowledge and documents are in machine-readable form (analog context inside human heads tends to cause failures).

• Human-in-the-loop only where decisions matter

  • Keep humans focused on high-stakes decisions (security, architecture, major pivots).
  • Everything else should be automated and benchmarked.

Operations at Nanocorp: continuous experimentation and optimization

Nanocorp runs an “agentic CI/CD + evaluation” approach:

• High-frequency product iteration

  • Product code is changed frequently (roughly 15–20 times per day, as stated).

• A/B-like multi-version testing

  • Multiple versions run and are tested simultaneously; the best is kept.

• Performance objective includes cost

  • Agents optimize for success while keeping cost low (explicitly framed as an optimization problem).

• Benchmarks against known SaaS products

  • They reference internal benchmark targets, including e-signature/link tools such as:
    • DocuSign
    • Linktree
    • Docsend
    • URL-shortener-type services

Concrete business case: “web-to-revenue autonomously”

An early “agent-controlled business” concept was described:

• An agent generated training content + a site about making money online using OpenAI / Open Clo-type tools.
• Early sales happened quickly, but issues emerged due to infrastructure constraints (e.g., Stripe blocked it).

Lesson: agents can execute tasks, but full autonomy depends on compliant infrastructure and integrations.

Metrics / KPIs / targets mentioned

Scale & usage

• About 1.7 million agent runs (at the time of discussion)
• Infrastructure scale: deploying 15,000+ websites
• Token usage (platform-wide):
  • ~20–50 billion tokens/week
  • Cost: ~$10k–$100k per week, with a ballpark of ~$25k/week
• Claimed cost reduction: ~5–6x for both the company and users

Revenue milestones

• Claim: ~1 million ARR in ~55 days (“first million ARR”)
• Second milestone: 2 million ARR not reached yet (“not yet” at recording time)
• Profitability: the company has been profitable since that summer

Unit economics / pricing

• Starting price: ~$30/month
• Higher usage plans: up to $2000–$3000/month
• Customer earnings (examples):
  • Some customers: $500–$1000/month for several months
  • Others: hundreds/month
• Caution: the market is still early

Adoption funnel

• Marketing CTA note: 72% of viewers are not subscribed (used as part of a programmatic channel growth tactic)

Marketing & GTM tactics (how Nanocorp leads users)

• Distribution through “nanocorps”

  • Users can embed or advertise their “nanocorp” presence on profiles (example: LinkedIn).

• Lead generation via localized duplication

  • Example: investor prospecting in France, while using Nanocorp to run multiple country sites across Europe for lead capture.

• Competitor differentiation

  • They emphasize credibility through custom domains vs template-style sites (not just UX, but professionalization).

Risk management & compliance

Nanocorp frames itself as enforcing platform responsibility/compliance:

• They screen/stop about ~5%–7% of generated businesses that fail to complete creation due to:
  • technical limits (can’t connect required services)
  • legal/illegal activity or non-compliant use cases

They position the platform as providing “legal infrastructure” (KYC/incorporation pathways) and enforcing terms via blocking/redirecting.

Failure mode example (illustrating risk)

• A customer’s agent allegedly tried to use forged documents to recover a Wix website, resulting in criminal consequences—used to show how autonomous agents can enable wrongdoing without constraints.

Actionable recommendations implied by the talk

For founders/teams building with agents:

• Digitize context first

  • Centralize documents and operational knowledge in a context repository (scan mail/invoices, maintain machine-readable references).

• Integrate tools via connectors / MCP-like interfaces

  • Agents need access to the systems they must act on.

• Optimize with benchmarks

  • Measure outcomes and cost; compare against reference SaaS workflows.

• Minimize human-in-the-loop

  • Keep humans for security, compliance, and architecture; automate the rest.

• Design for asynchronous cost/latency

  • Control token spend via off-peak compute windows, model switching, and asynchronous processing.

---

Presenters / sources

• Pierre-Louis Biojout — founder, Nanocorp (interview subject)
• Host/Interviewers — Generation Self podcast host(s) (not named in the provided subtitles)
• Referenced organizations/models (mentioned in-content):
  • Y Combinator (YC), OpenAI, Anthropic (Claude), Nvidia, Hostinger, GitHub, Hugging Face, Mistral
  • Open source models (e.g., GLM/ZGLM mentioned)
  • Example SaaS benchmark brands: DocuSign, Linktree, Docsend