Exhibit 062025 — 2026

Autonomous Systems

Software that runs the business.

An internal fleet of autonomous agents: 17 agents running marketing, sales, support and billing with human-in-the-loop approvals; an AI engineer that ships pull requests; trading bots executing on live markets.

fleet.primelabs.internal

Agent fleet · 17 active

marketing-agentdrafted 3 posts
sales-outreach12 emails queued
support-triageinbox clear
billing-dunningawaiting approval

✓ prime-agi · opened PR #214 — fix invoice rounding

→ tests passed (34/34) · awaiting human review

→ cost today: $0.41 / cap $5.00

Autonomous Systems2025 — 2026

MediumPython · FastAPI · Celery · Redis · Claude · GPT · open-weight models · Azure AI Foundry · Terraform

FormatDashboards · APIs · Live markets · Azure cloud

StatusInternal fleet, in operation

The thesis

The best way to sell AI automation is to run on it. Prime Labs builds autonomous systems for itself first — a company that markets, sells, supports and operates with agents — and brings clients the patterns that survived contact with reality.

The fleet

Seventeen agents cover the business surface: marketing and social publishing, sales outreach, support triage, onboarding, churn watch, reputation monitoring, billing and dunning, competitor intelligence, QA. Critical actions queue for human approval in a live dashboard. A multi-provider model layer routes each task by cost tier across Claude, GPT, Mistral, Llama and DeepSeek — the whole fleet runs on roughly $15 of LLM spend a month.

The engineer

prime-agi is an autonomous software engineer: scoped to a client, project and environment, it investigates services, edits code, runs tests and opens pull requests — with tier-gated permissions, an append-only audit log, daily cost caps, and a hard rule that it never merges or deploys itself. Work arrives from Jira, Linear, Azure DevOps and email.

The markets

Three generations of quantitative trading systems execute on live prediction markets — probability models over streaming exchange data, fractional Kelly position sizing, drawdown circuit-breakers and kill switches. No LLM in the hot path: statistics, discipline and risk management.

In the frame

What it does

A 17-agent fleet

Marketing, sales, support, onboarding, churn watch, billing — the business surface, run by agents on ~$15/month of LLM spend.

Humans gate what matters

Critical actions queue for approval in a live dashboard; everything else runs on schedule, 25 tasks deep.

An engineer that ships

prime-agi investigates, edits, tests and opens pull requests — tier-gated, audit-logged, never merging its own work.

Discipline on live markets

Probability models, Kelly sizing, drawdown circuit-breakers and kill switches — statistics in the hot path, not vibes.

The scope

What we delivered

  • 01Marketing, sales and support agents that draft, schedule, reply and follow up
  • 02An approval queue — anything critical waits for a human decision
  • 03An AI engineer that investigates issues, fixes them and proposes the change for review
  • 04Trading systems with strict risk limits and automatic kill switches
  • 05Dashboards showing every action taken and every cent spent

17

agents on the clock

24/7

the fleet never sleeps

100%

of critical actions human-approved

~$15

a month to run it all

Next exhibit

Kansa