Systems with Judgment

AI systems for music companies

Free 30-minute session

Second brain

Your company’s second brain. Explained through ours.

A second brain is an AI agent that holds your company’s context, does the recurring work on a schedule, and knows what to bring to a person. Ours is called Nova. It runs Futureproof Music School, and this page shows exactly how.

Nova reads every system the school runs, does thirty-plus jobs a day, repairs its own failures, learns every night, and brings the founder one decision at a time. Four people run like twenty.

Icon of a brain seen from above. The left hemisphere is an outline with folds, the right hemisphere is solid with circuit traces and coral nodes, joined by a coral line down the middle.

What it does for a company.

Not a productivity bump. A different size of company, with the same headcount.

Decide with the whole picture.

Revenue, retention, ad spend, support, and errors, read together every morning. Decisions get made on the full picture, not on whichever tab was open.

At Futureproof: one morning brief, one ask at most

Move faster than the competition.

The recurring work is done before anyone logs in. The team spends its day on what grows the business.

At Futureproof: 30+ jobs done overnight, every night

Get smarter every night.

Every mistake becomes a rule. A correction on Tuesday is standing policy by Wednesday, for every future run.

At Futureproof: 570+ memory entries, 30 operating rules

A chatbot answers. A second brain works.

A chatbot answers and forgets. A second brain is built differently.

It holds the context.

Rules, history, numbers, and who decides what, in files the agent reads before every run. Knowledge that survives staff turnover.

It works on a clock, not on request.

Nobody has to remember to ask. Thirty-plus jobs run every day on a clock. The team wakes up to finished work.

It knows what to escalate.

The founder reads three lines, not thirty pages. The detail waits in a report for whoever needs it.

Nova, from the inside.

No custom model. A frontier language model plus a very well-organised folder. That folder is the product.

  • 170+ named skills
  • 30 operating rules
  • 570+ memory entries
  • 110 canonical state files

Reads

Sources

  • PaymentsStripe, every dollar in and out
  • Course platformEnrollments, progress, activity
  • Ads and analyticsMeta, Google Analytics, Search Console
  • Email and communityThe inbox, the newsletter tool, Discord
  • Task boardTrello, both ways
  • The codebaseEvery app the school runs

Is made of

Nova

Operating document and rules

One instruction file Nova calls its soul document, plus thirty rules. Every run starts by reading them.

Memory, three kinds

What happened, what is true, how to do it. Six hundred entries, each losing confidence until re-verified.

Canonical state

One file holds the one true number for revenue, churn, and the rest. Nobody re-derives a figure by hand.

Skills, on a clock

170-plus named skills. Thirty-plus run daily, the rest weekly or on request.

Produces

Outputs

  • Reports and a dashboardWhere findings wait to be read
  • Task cardsFiled, deduplicated, closed when fixed
  • Email draftsIn the founder’s drafts, never sent
  • RepairsTo its own tools, verified before reported
  • Delegated codeBriefs handed to cheaper coding agents
  • One message a day, at mostTo the founder, who decides

Every night at 2:00am Nova reviews the day and writes what it learned back into memory and rules. Friday’s system is better than Monday’s.

Sources in, judgment in the middle, rationed output. The human stays at the end of the line by design.

A day in Nova’s life.

The team logs in to finished work.

Every job is a named skill with its own instructions and tools. If one fails, the 1:00pm run repairs it and escalates only what a person must do.

Nothing here reaches a customer. Emails are drafts, clips are staged for a person to post, and the founder gets a pointer to a report, not the report.

The rules that make it trustworthy.

An agent with access to your accounts is only useful if you can stop watching it. Each rule below was written after something went wrong.

  1. Verify before you claim.

    Another agent’s report is a hypothesis until it is checked against the source.

  2. Diagnose before you fix.

    Name and reproduce the cause, or the deliverable is an investigation, not a fix.

  3. One ask a day.

    Everything else waits in reports. Silence means handled, not forgotten.

  4. Drafts, never sends.

    Every customer email lands as a draft. A person who knows the relationship sends it.

  5. A broken tool stops everything.

    No silent fallbacks, no invented results. Report the break, fix the cause, then carry on.

  6. An autonomy charter.

    Three lists: ship and report, propose first, never. Payments, deletions, and pricing sit on the last. The first grows one approval at a time.

What Nova can’t do.

A second brain is powerful because of what it is not allowed to do.

It only remembers what a file remembers. Each run starts fresh. Continuity lives in files, which is also why the knowledge is inspectable.

It makes confident mistakes. Every rule above exists because it once did the wrong thing well. Scar tissue, and it works.

It doesn’t replace judgment. Nova surfaces. The founder decides. Human time goes to decisions, not to gathering the material for them.

“Memory records what was decided, not what was correct. My job is to pressure-test past decisions, not validate them.”

Nova, from its rules on using memory

How we’d build yours.

Nova took two years inside a live company. Yours won’t. The structure transfers.

  1. Map the week

    The recurring work and the recurring decisions.

  2. Write the brain

    The operating document, the rules, and the data connections, in one place.

  3. Schedule the first ten

    Monitors, reports, and drafts on a clock inside your accounts. Live in 4 to 6 weeks.

  4. Grow the autonomous zone

    Each class of fix earns standing authority as it proves itself. The nightly loop takes it from there.

See what your second brain would do first.

Thirty minutes, free. Bring the bottlenecks and the questions you answer by hand every week. We’ll tell you what a second brain would take off your plate first, or if it isn’t the right move yet.

Best for growing music companies with small teams, recurring work, and a founder reading too many dashboards.