The System of Work
A coordination substrate for work where humans and AI agents both participate as first-class actors.
The v4 thesis frames the System of Work as the explicit coordination layer beneath every organization: shared truth, authority, memory, intent, provenance, and the governance that lets human and AI actors contribute coherently as capability accelerates.
The argument
The long-form argument: every organization already runs on a coordination substrate. AI did not create that layer; it made the implicit version too fragile to ignore.
The field record
The companion record: a reproducible snapshot of what the substrate looked like in one repository after roughly six months of human and AI collaboration under production pressure.
The operating doctrine
The first-principles manifesto: patterns and strategies for building platforms, systems, and cultures that can evolve at the speed of agentic and embodied AI.
The coordination layer was always there
The thesis argues that work already runs on a substrate of shared truth, authority, memory, intent, and provenance. AI is the catalyst that makes this urgent, not the reason the substrate exists.
The substrate is a distinct layer.
Sprint plans, operating doctrine, governance, source truth, memory, and release process are not side documents. Together they are the operating infrastructure of work.
AI is the catalyst, not the premise.
The substrate matters with or without AI. AI makes every gap audible because agents cannot rely on unwritten social repair, social memory, or inferred authority.
Capability transfer becomes structural.
When product intent, architectural memory, ownership, constraints, and provenance are explicit, more actors can contribute across boundaries without extracting the organization from someone else's head.
Seven primitives
Identity
Who or what the actor is, including capability boundaries and earned standing.
Context
The shared truth all actors operate against right now.
Memory
What persists across sessions, teams, attempts, decisions, and failures.
Coordination
How concurrent action composes without hidden conflicts.
Intent
The purpose that travels with the work and can be checked against outcomes.
Governance
How actors are authorized to act, and how authority changes with evidence.
Provenance
What happened, why it happened, and on what basis the system should trust it.
The field record is about coordination, not only scale
The May 7 snapshot shows how work moved when humans and ordinary AI tools coordinated through shared files, release gates, verification, and source-tracked operating doctrine. The useful signal is the role shift: humans held intent, architecture, verification, and handoff while agents executed bounded slices in parallel.
First-class human and AI actors
Two engineers worked with Codex, Claude Code, Gemini CLI, and Replit in parallel, choosing tools by work shape rather than treating any tool as the system.
2 humans + 4 AI tools
collaboration mix
Christian and Vaibhav worked alongside Codex, Claude Code, Gemini CLI, and Replit.16.2K
reachable commits
Commit volume across all refs through the May 7 evidence window.1,411
merged PRs
GitHub-verified merged pull requests in the same snapshot window.110
active mainline days
Mainline commit days across a 194-day calendar span.Parallel orchestration signature
Activity clustered in simultaneous branches, sessions, and commit windows. The shape is orchestration, not single-keyboard serial work.
25
commits in one minute
Peak all-ref minute at 2026-04-15T23:28.1,018
commits in one day
Peak all-ref day on 2026-03-04.14
concurrent Codex sessions
Distinct sessions emitting events in the same UTC minute.23:00-02:00
peak commit band
Highest-volume hours clustered late evening through early morning; office-hour buckets were lower.Bounded autonomy
Top-level threads were orchestration lanes. Spawned agents usually received one boundary input and executed through completion.
64.20%
single-input subagents
Spawned Codex subagent sessions that ran from one coalesced input.83.29m
p90 autonomous window
Subagent autonomous-window duration at p90.179.15m
p95 autonomous window
Subagent autonomous-window duration at p95.115.27 days
paired-turn runtime
Aggregate paired-turn runtime across root and child Codex threads, overlaps not subtracted.1.96M
Codex tool calls
Recorded local Codex tool-call records in the snapshot corpus.Governance that makes speed legible
The autonomy sat inside repo-owned procedures, verification records, ADRs, and a compiled control plane shared by humans, agents, and CI.
127
workflow specs
Source procedures under the System-of-Work domain corpus.73
governed commands
Slash-command surfaces projected from workflow specifications.858
verification reports
Packet verification records retained in the work corpus.159
ADRs
Active and archived architectural decision records.279,656
control-plane lines
Rust implementation lines under tools/fireweed, invoked by humans, agents, and CI.One construction under real operating pressure
The evidence is not a benchmark. It is a field record of one startup construction: two human engineers and parallel AI tools building a production-shaped platform, product surface, infrastructure, and governance system through a file-first substrate.
Selected observations
633K
implementation lines
Non-generated Rust, TypeScript, TSX, and Python implementation surface.16 + 5
services and apps
Backend services plus web, commerce, mobile, and public product surfaces.30
proto files
Contract surface across 26 top-level domains, with generated stubs drift-checked.87
migrations
Forward-only Postgres migrations across the data and coordination planes.2,724
test functions
Measured Rust and TypeScript test functions, alongside Bazel and E2E targets.127
workflow specs
Procedure sources projected into governed skills, agent surfaces, CLIs, and CI.1,670
mainline commits
Mainline history over 110 active commit days inside the snapshot window.5
live endpoints
Public health checks responding for Eigenly, Fireweed, Wishflower, API, and assets.Production-shaped output.
The snapshot includes services, apps, contracts, migrations, infrastructure, tests, and live health checks rather than a thin demo surface.
Parallel orchestration.
Commit bursts, concurrent Codex sessions, spawned agents, and multi-client telemetry show work decomposed and recomposed across parallel lanes.
Humans moved up the chain.
Human input clustered around intent, architecture, planning, review, and handoff while agents executed bounded implementation work inside the substrate.
Governance made autonomy viable.
Procedure specs project into skills, agent surfaces, repo-owned binaries, verification reports, and CI gates that preserve provenance.
The interpretation stays bounded.
The page summarizes what the snapshot shows; the PDF preserves the source record and method disclosures for readers who want the full detail.
Read the originals
The web page is an orientation layer. The PDFs remain the v4 long-form thesis, the method-preserving evidence record, and the first-principles operating manifesto.
The System of Work
The v4 long-form thesis for the coordination substrate, why it is not only an AI problem, and how explicit work systems transfer capability across roles and agents.
Download PDFThe Evidence
A structured Eigenly and Fireweed field record with metrics, operating signals, production proof, and reproduction commands.
Download PDFPrinciples for Building in the Era of AI
The public manifesto for first-principles operating doctrine: how platforms, systems, and cultures stay adaptable while human and AI collaboration accelerates.
Download PDFThe evidence is a single repository and local-machine telemetry snapshot from May 7, 2026. It is an observation of one construction, not a universal benchmark, customer-adoption proof, or claim that the same substrate transfers unchanged to every organization. The principles manifesto is included as part of the System of Work source set.