How One Developer Runs a Multi-Project Business With 17 AI Agents

Overview

A solo entrepreneur has revealed how he operates a multi-project business using a team of 17 distinct AI agents. Blake La Pierre, a computer scientist with a background in networks and distributed systems, treats his one-person enterprise as a complex infrastructure problem rather than a traditional consultancy.

By delegating specialized tasks to narrow-focused software assistants, he manages everything from crypto trading bots to published puzzle books. The approach allows a single human to run experiments and operations that would typically require a larger staff.

The Distributed System Approach

La Pierre’s methodology stems from his technical background. He realized that acting as a sole proprietor meant wearing too many hats simultaneously—bookkeeper, researcher, marketer, and tax specialist. Rather than relying on a single general-purpose chatbot for all queries, he adopted the principles of distributed systems.

He split the workload into separate agents, each with its own memory, context, and specific job description. This isolation ensures that one agent handling sensitive financial data does not accidentally influence the context of another agent working on creative writing projects.

That separation turned out to matter more than I expected. One agent knows everything about my puzzle books and nothing about my taxes.

This architecture prevents context bleeding, a common failure point in simpler AI setups. Each assistant operates within a defined boundary, allowing the human owner to oversee multiple ventures without cognitive overload.

Operational Portfolio

The fleet of agents is divided into four main categories: publishing, trading and markets, money hunting, and business administration.

Publishing and Creative Assets

The Books agent manages the entire lifecycle of La Pierre’s Amazon KDP puzzle books. Source files, print layouts, and listing drafts are stored in a git repository. Two titles, The Thief Stayed the Night and Frostwood Express, are currently live in his "Frostwood" cozy winter series.

Additionally, twelve more titles—including logic puzzles, cryptograms, and an advent puzzle book—are built and print-ready but awaiting publication. Another agent powers Streaming Audio, a 24/7 YouTube live stream of generative music synthesized in real-time using numpy. Viewers vote via live chat on the next musical style, mood, and instrument.

Trading and Market Intelligence

La Pierre runs a suite of crypto market-making bots on Coinbase and Kraken. These include a "ladder" bot and Avellaneda-Stoikov market makers. The code for these strategies is public.

Two agents support this side of the business:

Trading strats: Designs strategy parameters.
Trade bot status: Monitors live performance, diagnoses losses, and writes change briefs.

In one notable incident, an agent detected a $1.14 loss at 3 AM. It logged the issue to a minor bookkeeping error where market orders bypassed internal tracking. The agent then drafted a fix, including a dry-run plan and rollback strategy.

For manual trading, a separate agent built an alert-only breakout scanner for Kraken prop trading. It monitors 67 perpetual markets hourly, printing entry and stop-loss points without placing orders. A Options scanner also hunts for asymmetric payoff opportunities in leveraged-ETF options.

Revenue Optimization

Three agents focus on direct revenue generation:

Bonus harvesting: Scans bank sign-up bonuses and brokerage promos.
Referral bonuses: Tracks 36 active programs, valuing points conservatively.
Money making: Researches trading competitions and incentive programs for US residents.

These tools help identify deadlines and eligibility requirements, ensuring no promotional opportunity is missed.

Business Administration and Tax Compliance

The remaining agents handle the administrative backbone of the sole proprietorship. The Sole proprietor Odoo bot maintains the open-source ERP system, managing timesheets and reports.

The US tax code agent answers statutory questions by pulling directly from official sources. When the primary government website was down for maintenance, it successfully retrieved the Title 26 XML release from an Internet Archive capture rather than guessing or using cached data.

Other agents research small business capital options, tag funding claims by verification status, and monitor an internal consulting channel for ad-hoc tasks. A generalist agent handles overarching questions, maintaining a big-picture view of the entire operation.

Results and Philosophy

La Pierre describes the setup as a "build-in-public" experiment. He notes that while the system is not entirely autonomous—human oversight remains necessary—it significantly increases throughput. The ability to run multiple bots and publish books simultaneously demonstrates the scalability of AI-assisted solopreneurship.

The model prioritizes reliability and separation of concerns. By giving each agent a narrow job, failures are contained, and the overall system remains stable even when individual components encounter issues. This newsletter series documents the real numbers, including losses, to provide a transparent look at what works and what does not in this emerging operational model.