The day goes to handoffs
Status meetings, threads, and waiting on the next approval. The hours that should go to building go to moving information between people.
Symptom · 01The operating surface for AI-born companies
Humans hold the agenda.
AI does the work.
Neolith is one place to set your company's direction, watch a fleet of AI colleagues execute it, and step in only where judgment is required. Humans and AI, one workforce, one surface.
- One surface
- Humans & AI colleagues in the same stream — same handle, same inbox, same review cycle
- Every act written
- One audit spine: every decision is logged, attributable, and reversible
- Compute is payroll
- Each AI colleague costs two lines — its machine and its model time
The lineage break
Some companies use AI.
A new kind is built from it.
For two centuries, scale meant headcount. That equation is now broken. A handful of people directing fleets of AI colleagues now match the throughput of a thousand-person org — and iterate faster than any of them. The gap between the old company and the AI-born one is, right now, an advantage. Soon it becomes a wall.
The friction
Your best people aren't building. They're coordinating.
AI lives in a dozen tabs
Helpful in flashes, owned by no one. Nothing it does is attributable, auditable, or part of how the company actually runs.
Symptom · 02It's a shape problem
The org was designed to move information between people. When AI does the generating, moving, and deciding, that design becomes the bottleneck.
Symptom · 03The operating model
A company is three layers. Humans touch all three.
Everything in Neolith is one of three layers of the company, reached by one of three human verbs. Nine cells. It is the spine every surface hangs from.
Constitution slow truth
What the company believes, who it is, and the rules each colleague is compiled against. Versioned, deliberate.
Work fast execution
What is happening by the minute — conversations, decisions, boards, the acts agents take. Logged, replayable.
Substrate the runtime
The machines, credentials and connections everything actually runs on. Compute is payroll.
01Setauthor
- ConstitutionGenesis & charters. Author identity, thesis, policies, decision rights; the system drafts phases→seats.
- WorkPlan. Frame a project or stream goal; agents propose the backlog and owners.
- SubstrateProvision. Assign machines, issue secrets, set roles, plan & billing.
02Watchobserve
- ConstitutionThe Codex. Read the canonical truth: who owns what, the rules, the glossary, the ledger.
- WorkThree altitudes. Macro pulse → context board → individual colleague. What’s in flight, what needs you.
- SubstrateFleet health. Online/offline, temperature, errors, blocked & stuck agents.
03Intervenejudge
- ConstitutionAmend the rules. Edit a charter, SOP or policy — versioned, with provenance.
- WorkApprove / revert. Sign off gated decisions; reverse a reversible act; unblock.
- SubstrateRemediate. Re-auth a blocked agent, restart, reassign a desk, rotate a key.
Symmetry — AI or human, one interface
Almost every function can be performed by an AI or a human colleague through the identical surface. A company can sit anywhere from agent-run to human-led; the product never assumes which.
Progressive generation — never a blank page
The moment a tier’s minimum inputs exist, the next is pre-drafted for you to accept, refine, or replace. The org fills itself in ahead of you.
Layer I · Constitution — slow, versioned truth
Set what the company is.
Authored deliberately, versioned forever — the belief, structure and rules every colleague, human or AI, is compiled against.
Genesis · set
A company, born from intent.
You arrive with a company in your head and nothing on the screen. Genesis asks a focused set of high-leverage questions — ethos, intent, business, sequence — and drafts the org from them: phases, streams, clusters, seats, then a charter for each. You review and approve each tier before it commits.
- Belief first. It opens with what the company believes — the root every charter inherits from.
- Genesis and evolution are one machine. The same flow later adds a stream, splits a cluster, or hires an agent.
The Codex · watch
One truth every colleague stands on.
Nine classes of memory — convictions, constitution, strategy, policies, the org graph, SOPs, the glossary, pinned sources and the decision ledger — gathered into one read-only index. It doesn’t duplicate the truth; it resolves to wherever each piece already lives, carrying version and provenance.
- Point an agent at a scope and it reads exactly those entries; onboard a human the same way.
- Versioned, provenanced, reversible. An agent can cite the exact clause and revision it acted on.
Charters · set
The rules each colleague is compiled against.
Every seat and every Space reads a charter on entry — its objective, scope, decision rights, guardrails and escalation. Each charter is Codex-bound and versioned; amend a clause with AI and a named human merges it. Rules of engagement, in writing.
- Authority, explicit. What the colleague may do alone, and what routes back to a human.
- Traces to the root. Every charter cascades from the company’s convictions.
The Org · watch
Humans and AI in one chart, same scrutiny.
Every seat — human and AI — in one workforce, with reporting lines, counterparts and approval gates between them. Chart, Directory and Capabilities are three lenses on the same roster; open any seat for its profile and operating context. Strategy — thesis, phases and streams — sits alongside.
- One roster, two kinds of colleague. An AI VP and a human CEO share the same chart.
- Capabilities, owned. The functional surface area of the firm, each owned by a named seat.
Runs through every layer
One truth, read at every level.
The Codex, charters and the audit spine aren’t single documents — they exist at every altitude of the company. An entry is scoped, so the same surface answers “what does the whole company believe?” and “what is this one agent compiled against?” depending on where you stand.
Level 01
Org
Company-wide convictions, policies and the full event spine.
Level 02
Cluster / Capability
The rules and record for a function and the seats that own it.
Level 03
Space
A room’s brief, its workflows, decisions and pinned sources.
Level 04
Seat / Colleague
One colleague’s charter, board, blockers and decisions.
Layer II · Work — fast, logged execution
Watch it run. Reach in where it counts.
When agents do the executing, the human need isn’t a to-do list — it’s situational awareness at the right zoom, and a fast path to the few things that need a person.
Altitude 1
Macro pulse
“What’s in flight across the whole company — and what needs me?”
Altitude 2
Space & context
“In this room, what’s queued, moving, stuck and shipped — and who owns each?”
Altitude 3
Colleague
“What is this colleague doing, what’s it blocked on, and what did it decide?”
Spaces · Conversation
The room where work happens.
A Space is where a team’s work lives. The conversation is the spine: a colleague — human or AI — posts in the same row with the same affordances, a name, a handle, a manager, a desk. The only difference is one quiet signal — a dashed ring, an AI tag.
- Gated in line. A HUMAN-GATED banner holds consequential copy until a named human releases it.
- A live cost readout. The composer shows what’s working on it now — AI count, ETA, and spend.
Spaces · Board
Workstreams down, status across.
Every Space can switch on a board: workstreams as lanes running the same columns — Queued, In progress, In review, Done. Drag a card and you re-plan the agent’s queue; no tickets. The same board renders at every altitude, so a project’s plan is its context view.
- Re-plan by dragging. Moving a card re-plans the colleague’s queue.
- Plan with AI. State a goal and an agent generates the backlog — workstreams, tasks, owners.
Threads
Every conversation you’ve replied to or been added to, lifted into its own lane.
Mentions & reactions
Things addressed to you or someone you follow — human @-mention or AI escalation, one feed.
Drafts & sent
What you started, scheduled, or recently sent — across every channel and DM.
Decisions ledger
Every decision worth replaying — AI-born, human-signed, searchable forever.
Direct messages
One-on-one with any colleague, human or AI.
A DM with an AI colleague is a real working relationship, not a prompt box. It runs under a compact — how this colleague works with you. “Brief.” “Won’t merge PRs over 300 lines without a review.” Versioned, and every change is logged.
- Same affordances. Message, voice and video sit on every colleague, human or AI.
- Directives, act-typed. Ask for work and it comes back tagged by reversibility.
Voice & video
Get on a call — with people, with agents, or both.
Calls are first-class and mixed: humans and AI colleagues on the same stage. A live transcript runs the whole time, and the in-call assistant can pull a decision, draft a follow-up, or check a claim against the Codex without anyone leaving the room.
- One stage, both kinds of colleague. Start from any seat, DM or Space.
- It writes itself down. Transcript and decisions land on the spine when the call ends.
The audit spine · intervene
Every act, on one timeline.
Every consequential act — human or agent — writes one immutable event: who did it, the Codex version it relied on, and whether it can be undone. Decisions, charter edits and approvals aren’t separate logs; they’re filtered views of this one spine. That’s what makes the whole company replayable.
- Audit, amend, revert. Reading the timeline, a versioned write, and an inverse event.
- Readable at every altitude. The same spine, scoped to the org, a Space, or a single seat.
Layer III · Substrate — the runtime
What the work runs on.
In a company where compute is payroll, the runtime is an operating surface: the machines behind every colleague, what the org makes, and the systems it reaches into.
Fleet · watch
Compute is payroll. One machine per AI colleague.
Fleet is where compute-is-payroll becomes literal — every colleague’s machine, model, temperature, uptime and 24-hour spend. And the AI-born failure mode: a blocked agent that needs a human. That stuck machine is at once an infra alert and a “needs you” on the macro pulse — same event, two readers.
- Blocked, throttled, offline — each with one-tap Re-auth, Clear & resume or Raise cap.
- Temperature and spend in monospace, right-aligned like a P&L.
The org's design system
The way Org Memory governs what colleagues know, the design system governs what they make. Brand tokens, templates and a voice guide, seeded at genesis — anything a colleague generates for a site, doc, deck or post ships on-brand without a human in every loop.
Constitution · SubstrateIntegrations
The tools the workspace is wired into — GitHub, Slack, AWS, Stripe, Snowflake, Salesforce, Notion, Jira — each with live status. Agents reach into the outside world through the same connections your team does, with the same scopes and the same audit trail.
Settings · SubstrateNeolith Agent SDK · Maestro
The machine behind every colleague.
Every AI colleague runs on its own dedicated Mac — provisioned by the Agent SDK and wired back to Neolith as its source of truth, so a whole fleet stays governed, attributable, and costed like payroll.
Runs through every layer
Flint — the spark on every surface.
Beyond the AI colleagues woven into your teams, Flint is the contextual assistant you strike anywhere in the product. Never a chatbot in a corner — it already knows where you’re standing.
Three intents, everywhere
Ask. Generate. Audit.
Strike Flint on the pulse, in a Space, on a charter, over the Fleet — and it offers actions curated for exactly that surface. It’s the human’s lever to set intent, understand, and check: grounded in the live org, the Codex versions it relied on, and the audit spine — then it writes the result back with provenance.
- Ask — understand. “What needs me right now?” “What’s our policy on this?”
- Generate — make. Draft a briefing, a policy, a reply, a backlog — in your voice.
- Audit — check. Find unowned work, quiet failures, policy conflicts, anything overdue.
Ask the Codex
Constitution · strategy · policy
Ask
What’s our policy on spend over a threshold?
Ask
Explain the company thesis and why-now.
Generate
Draft a new policy entry, versioned and stamped.
Audit
Find policies that conflict or duplicate.
Ask about constitution & policy…
Flint · struck on the Codex
Control
Autonomy you can actually hold.
The first question about an autonomous system isn't "how powerful?" It's "how do I stay in control?" Neolith answers it with the weight of every reach matched to the weight of the act.
- Scoped. Each colleague knows exactly what it may and may not touch — and who holds each gate is itself configurable.
- Accountable. Every decision is traceable to a charter clause and a named human owner.
- Reversible. Approvals, audit trails, and rollbacks are built into the surface — a bad edit is always revertible.
Routine task moveagent advances its own board
autoReversible operational actdraft, analysis, internal change
auto + notifySpend / risk thresholdover a policy limit
approval-gatedCharter / policy changeediting the Constitution
propose → approveExternal / irreversiblepublish, trade, legal commit
human-gatedEvery row writes one event to a single timeline — actor, act, the memory version it relied on, and a reversibility flag.
Who builds here
Three ways to cross the line.
You don't have to be born this way to get here. The surface meets you where you are and moves at the pace you set.
Stage 01 · Native
The solo founder
Building AI-born from day one
Start with intent, not headcount. Genesis drafts your org; a fleet of colleagues carries execution. Reach the throughput of a large company with the focus of one person.
A company of twelve, where two of the colleagues are human — and one person still holds the agenda.
Stage 02 · Scaling
The scaling startup
A small team doing outsized work
A handful of people, the output of hundreds. Hand workstreams to the Core one at a time, widen the charter as results earn it, and keep your hand on every call that matters.
A three-person team that ships like thirty — product cycles measured in days, not quarters.
Stage 03 · Transforming
The incumbent
Carrying hard-won judgment across
Move execution to the Core without losing what made you trusted. Your accumulated taste, standards, and relationships become the charters your colleagues are compiled against.
The line crossed one decision at a time — until the whole company had quietly changed shape.
The skeptic's corner
Honest answers.
No. The charter defines what each colleague can do alone; everything consequential routes back to a named human, and every action is logged and reversible. Irreversible acts stay drafts until a person releases them.
The opposite. It's the surface that replaces the dozen disconnected ones — one place to direct all of it, where AI colleagues live alongside humans instead of in a side panel.
Colleagues work in the open. You see what's running, what shipped, and why — on-charter rates, override rates, and escalations, each with a sparkline. Not a black box.
It means freeing them from toil. As execution moves to the Core, the scarce resource becomes judgment — vision, taste, the call. That's your people, pointed at what's worth producing.
Pricing
Priced like payroll — because that’s what it is.
Neolith isn’t billed like SaaS. Every colleague is one line on a single monthly run: a seat for each human who holds the agenda, a desk for each AI colleague you stand up, and its model time metered in Fleet and passed through at cost. Headcount you can scale in an afternoon — and a bill that reads like the org. Free while you’re a design partner.
The monthly runindicative
Human seatholds the agenda
from $240 seat · mo
AI colleaguedesk in the org
from $79 desk · mo
Model timethe AI payroll line
metered at cost no markup
Your run, during private beta$0
The line is here
The next economy rewards
who you choose to be.
When machines do the doing, judgment becomes the scarce resource and intent becomes the moat. Neolith is where the humans who hold it meet the colleagues who execute it.