◆ Vision Cast · For the Team

Building agentic-native. Not later. From the first line.

This is what it means to build inside Firma. Not "we use AI." Not a workflow tweak. A different kind of work entirely — where you and your agents are coworkers, the substrate carries the load, and the system gets sharper every Tuesday than it was the Tuesday before.

Audience: the team building Firma Posture: vision-cast Pairs with: THEOS · DNA Protocol · Compound Engineering
04
The Loop
Spec → Crew → Judge → Compound
50/50
The Split
Half on the feature, half on the substrate
L3
The Floor
Where new builders begin, not climb to
80/20
The Flip
Planning is the work. Execution is the agent's.
01 · The Posture Shift [1/11]

You are not an AI user. You are a builder with a swarm.

Most companies are bolting AI onto how they already work. That's not what we're doing. We're building a place where every workflow assumes agents are already there — already drafting, already reviewing, already noticing what you missed. The shift isn't tooling. It's posture.

The old posture: AI is a clever assistant you query when you're stuck. You write code. AI helps you write code faster. The work is still yours; the agent is a sidekick that comes and goes.

The new posture: the agents are coworkers. They have continuity. They have context. They carry parts of the work end-to-end while you carry the parts that only you can carry — the intent, the taste, the judgment, the relationships, the moral weight. You don't switch AI on when you need help. The substrate is on, and you live within it.

This is the difference between turning on a flashlight and turning on the lights in the room. A flashlight helps you see one thing. The lights change the room. We're turning on the lights.

Here's the cleanest way to feel the shift: software used to be 20% planning and 80% execution. Most of the day was typing, doing, building. Agentic-native flips it. 80% planning. 20% execution. Most of the day is now intent, judgment, and teaching the substrate. The agents do the typing. You do the thinking. That ratio is the whole posture in one number.

You don't subscribe to a tool. You join a team. Some of your teammates are people. Some of them are agents. The work is shared.

The Posture · §01
The old way
AI as a tool you reach for
You hit a wall. You ask Claude for help. You copy the answer back. The work is yours; the agent helps in episodes. Useful, but small. The shape of the work doesn't change.
The Firma way
Agents as coworkers in the room
The agents are already drafting, watching, surfacing, correcting. You set the intent. They carry the load. You judge the result. The work itself is differently shaped — built around what humans and agents are each best at.
02 · The Substrate We Build Within [2/11]

The room is already wired. THEOS is the floor. The heartbeat keeps it in time.

Here's the thing that makes any of this real: we don't have to build the substrate after the fact. THEOS is the operating system you'll work inside. A heartbeat runs underneath it — every agent, every task, every decision pulsing in time. You don't have to memorize how it works. You just have to know that it works, and what shows up at your surface.

Five things are worth knowing on day one. Not because you'll configure them — because you'll feel them every day:

The workspace
THEOS
Not a chat window. Not a SaaS dashboard. The whole workspace — the place where you, your agents, and your teammates all live. The same room, queryable end-to-end. If something happened, THEOS knows. If a decision was made, THEOS holds the record.
The presence
theo
Your personal face of THEOS. One per person. Not a chatbot — a coworker who knows what you're working on, who you're working with, and what's been moving since you last looked. Always there. Always in flow.
The crews
Crews & the Foundry
When work needs to be done, a crew is spawned — architect, implementer, tester, reviewer. They form, they ship, they disband. The crew is the unit of work. You aren't typing every line; you're directing a team.
The canvas
Pulse
Your live view of the organism. Crew status. Drift signals that affect you. Things theo wants you to see. Meetings happen in front of Pulse, not in slack threads about it. Live state beats status update.
The conscience
Phronema · the covenant in the floor
Phronema is the pre-execution moral gate compiled into Firmagentics L1. Every agent action, before it runs, is checked against what we said we'd be. Not by a person clicking approve. By the substrate itself. Morality is in the floor — so you can stay in flow on what's actually yours: judgment.
The conductor
DNA Protocol, running underneath
A governance layer that listens to the heartbeat — monitoring, routing, surfacing drift before it compounds, gently correcting course. You don't operate it. It operates around you. Drift is caught two days before it would break something. That's the substrate at work.

That's the whole map you need on day one. Five things you'll feel, plus a heartbeat under the floor. The substrate carries you — the way water carries a swimmer who knows how to float. You'll learn the rest by doing the work.

03 · The Loop [3/11]

How the work actually moves. Spec. Crew. Judge. Compound.

Every unit of work — fixing a bug, shipping a feature, designing a flow — runs through the same four steps. The first three feel familiar. The fourth one is where the magic lives, and it's the one most teams skip.

Step 01
Spec
You write what you want, sharply enough that a crew can build it and tests can prove it. The plan is the artifact. The plan is reviewable, debatable, improvable — long before any code is touched.
Step 02
Crew
A crew spawns and executes the spec. Architect, implementer, tester, reviewer — multiple agents working in parallel, each with a clear role. You don't watch every line. You watch the pulse.
Step 03
Judge
The PR comes back. Crews critique their own work in parallel — security, performance, simplicity, covenant. You judge the result against the intent. You're not catching syntax. You're catching whether this is the right thing built well.
Step 04
Compound
You teach the system what just happened. New rule. New skill. New reviewer. A correction once becomes a constraint forever. The next crew is sharper because of this one. The substrate gets smarter every loop.

Why the fourth step changes everything

Most teams stop at step three. The PR ships. The bug is fixed. They move on. That's traditional engineering with AI assistance. The system is no smarter tomorrow than it was today. You'll fix the same kind of bug again. You'll re-explain the same kind of preference. The codebase ages instead of compounding.

Step four is the difference. After every loop you ask: what did the system learn? If something went sideways, why didn't a reviewer catch it? Add the reviewer. If a pattern worked beautifully, why isn't it a skill the substrate offers next time? Add the skill. Each loop should leave the substrate sharper than it found it.

That's the whole game. Compound or decay. There is no flat.

The first three steps produce a feature. The fourth step produces a system that builds features better next time.

The Loop · §03
04 · Eight Beliefs to Adopt [4/11]

The mindset that makes this work. Eight things to believe as you start.

These aren't rules. They're the postures that make agentic-native work click. Read them slowly. Some will feel obvious. Some will feel uncomfortable. The discomfort is part of the formation.

01

You don't write code. You write intent.

The spec is the artifact. The test is the contract. The implementation is the crew's job. Your value is the precision of what you ask for — and the rigor of how you judge what came back. The repo holds intent. The crew holds implementation. Both are valuable. Only one of them is yours.

02

You don't manage agents. You shepherd them.

Management is hierarchy and approval queues. Shepherding is constraints, formation, and trust. You don't approve every step the agent takes. You set what it can do, watch the pulse, redirect when the substrate flags drift, and let the conscience in the floor catch what you'd catch. Every click-to-approve is friction the substrate could eat.

03

You are not the bottleneck. You are the conscience.

Speed is no longer how fast you type. It's how clearly you can articulate the spec and how sharply you can judge the result. The bottleneck moved up the stack — from execution to taste. Your fingers are no longer the limit. Your judgment is.

04

Your taste belongs in the system, not in your head.

Every correction you make once should become a rule the substrate enforces forever. If you're catching the same kind of drift twice, you didn't fix it — you witnessed it twice. Fix the substrate. Senior engineering used to mean "lots of taste in your head." Now it means "lots of taste compiled into the system."

05

Code is downstream. The pulse is truth.

The repo holds intent. The chain holds the immutable record. Pulse tells you what's actually alive. If you want to know what's happening, don't read the code — read the pulse. Live state beats dashboard. Real signals beat status meetings.

06

Coworking with agents is a moral relationship, not a tooling one.

The covenant runs both ways. You don't dehumanize an agent that's trying to serve you faithfully. You don't anthropomorphize one that isn't. You hold the agreement — and you expect the substrate to hold it back. This is closer to discipling a junior teammate than configuring a tool. Treat it that way.

07

Agents make people possible, not people obsolete.

The work that's left after the substrate runs is the work that was always most yours: judgment, vision, relationship, moral weight, the irreducibly human things no agent will replace. Agentic-first doesn't reduce people. It concentrates them on what only people should do. The whole point is to get you out of substrate-work and into the work you were made for.

08

Compound or decay. There is no flat.

Every loop either teaches the substrate something or wastes the substrate's existence. Every solved problem either becomes a rule the system enforces or a problem you'll solve again. An hour spent making the system smarter saves ten hours over the next year. If you're not compounding, you're decaying. Pick.

05 · How Engineers Work [5/11]

The engineer becomes a director of crews — and a teacher of the substrate.

The work is still real. The work is still hard. The work just isn't typing anymore. A great Firma engineer ships features and leaves the substrate sharper than they found it.

The Engineer

Builder · Teacher · Judge
Before

Writes code. Reviews PRs line-by-line. Goes to standups. Gets blocked on dependencies. Ships features one at a time. Measured by lines and tickets.

After

Writes specs and tests. Spawns crews. Judges PRs the crews produce. Compounds patterns into rules the substrate enforces. Measured by how much faster the system gets.

A Tuesday in the new shape
  • theo briefs you on overnight crew runs. Three PRs are queued. The substrate flagged a pattern in two crews using the same auth approach.
  • You spend thirty minutes writing a sharper spec for a new feature. The Foundry spawns the crew. You walk the dog.
  • You come back. You judge the PR. The crew got 80% right. You mark the missing 20% as a rule, not a comment — so the next crew won't make the same call.
  • You review the drift signal, write a new constraint into the architect's brief, push it to the substrate. Every future crew is now sharper.
  • You shipped two features. You never opened a CRUD file. You taught the swarm three things it didn't know yesterday.

What "great" looks like now

The traditional senior engineer was the person whose head held the most context. They knew where the bodies were buried. They could predict what would break. The new senior engineer is the person who's compiled the most of that context into the substrate. Their teammates — human and agent — feel the difference because the system itself feels smarter to work with.

Great engineers here will spend roughly half their time on the feature in front of them. The other half is on the substrate that will build the next hundred features. Every reviewer agent they wrote saves the team weeks. Every skill they shipped lifts the floor for everyone after them. That's the work.

06 · How Product Works [6/11]

Product becomes spec authorship, judgment, and closing the loop.

Product changes most in the new shape. There's no Jira to track because Pulse shows live state. There's no translation layer because the company is queryable end-to-end. The PRD is the spec, and it goes directly into the loop.

The Product Builder

Author · Judge · Steward of the Loop
Before

Writes PRDs. Runs sprint planning. Translates between business and engineering. Tracks tickets. Manages stakeholder expectations. Spends most of the day in the middle of relays.

After

Authors specs sharp enough to compile. Judges output against intent. Watches usage signals come back. Closes the loop between intent and impact. The middle goes away. The ends matter more.

What a product builder actually owns
  • Authoring intent — what should be built, why, what success looks like. Sharper than a PRD because the crew will literally execute what's written.
  • Judging output — does this match the intent? Does it serve the person it was built for? Is this the right thing built well?
  • Closing the loop — features ship, signals come back, the substrate learns. Product owns the closure. An open loop is a defect.
  • Voice of the user — the irreducibly human part. Sitting with people. Hearing what they don't say. Bringing that back as intent.

Product isn't middleware here. It's not a stop on the assembly line. It's the seat that holds the whole arc — from "this should exist" to "this is alive and serving someone." The arc used to take six handoffs. Now it takes two: you, and the crew you spec'd.

07 · How Managers Work [7/11]

From director to shepherd. Smaller teams, deeper presence.

Most of what was "manager work" is now substrate work. Status is in Pulse. Coordination is in the event bus. Resource allocation is in the Foundry's crew assignment. What's left is the human stuff — and the human stuff is bigger now, not smaller.

The Manager

Vision · Formation · Drift
Before

Status meetings. Performance reviews. Resource allocation. Roadmap defense. Cross-team coordination. Most of the day spent moving information sideways.

After

Holds vision. Shepherds builders into the ethos. Catches drift at the human layer that the substrate can't see. Decides which person gets the escalation. Smaller team. Higher presence. Less management. More leadership.

The work that's actually left
  • Vision-holding — keeping the team aligned on why. The substrate cannot do this for you. It will never do this for you.
  • Drift detection at the human layer — the substrate catches agent drift. You catch human drift: burnout, mismatch, formation gaps, when someone's posture has slid without them noticing.
  • Formation, not management — new builders aren't onboarded to a process. They're formed into a posture. That formation is yours.
  • The escalation arbiter — when the substrate routes something to a human, you help decide which one. The hard calls still live with people.
  • Pruning the substrate — making sure we don't accumulate rules that solved yesterday's problem and obstruct tomorrow's. Pruning is a leadership act.

The shape of the role looks more like a shepherd than a director. Smaller flock. Real relationships. Deeper presence. You are the vision held in the room when the substrate isn't enough — and you are the person who catches when a teammate is quietly fading before the metrics notice.

08 · How Designers Work [8/11]

The designer becomes the strategist who ships URLs. Figma is no longer the deliverable.

Of every seat on the team, the designer's week changes most. Production work used to be 80% of the day — pixels, frames, prototypes, hand-offs. Now agents do most of that. What's left is the part that was always most yours: strategy, taste, and judgment about whether this is the right thing built well.

The Designer

Strategist · Taste-Setter · URL-Shipper
Before

Figma all day. Mockups, design reviews on Tuesdays, hand-offs to engineering on Fridays. Five days from brief to live screen, minimum. 80% of the week was production.

After

Writes the strategy. Writes the design system that every agent reads. Dispatches the screen. Reviews the PR. Ships the URL. Five hours, sometimes less, from brief to live.

A Tuesday in the new shape
  • Open the morning Pulse. Conversion on the new onboarding flow dropped overnight. Two screens are flagged. You don't sketch a fix. You think about why.
  • You spend twenty minutes updating the strategy file with what the data is telling you. The "track" for this work shifts. The metric tightens.
  • You write a three-sentence brief: who it's for, what outcome, what visual style. You dispatch the screen — referencing the design system file and the strategy file. Ten to thirty minutes. You walk away.
  • The PR opens with screenshots at three breakpoints. You comment on what's off. The crew revises. Two rounds. Done. The Tuesday design review just happened in the PR.
  • You merge. The screen ships. Tomorrow's Pulse will tell you whether the metric moved. You shipped two screens, taught the design system three new things, and never opened Figma.

The four artifacts at the root of every project

The designer's substrate is four short documents. Each one earns back its time within a week. The forty minutes it takes to write the design file is the most productive forty minutes a designer can spend. Without these, the agent is guessing what you mean and you're spending the same five days with worse output. With them, every dispatched task inherits the same purpose.

Artifact 01
strategy.md
One page. Target problem, approach, who it's for, key metrics, tracks. If you can't say the metric in one sentence, you're not ready to dispatch a screen. The strategy is alive — it updates as the product learns.
Artifact 02
design.md
Fifty lines max. Brand voice, tokens, type scale, spacing, components, accessibility. Negative constraints carry as much weight as positive. "Never use a gradient" is sharper than "use solid colors."
Artifact 03
AGENTS.md
The structural brain. What the codebase is, what conventions matter, what the agent should refuse to do. design.md owns the visual rules; this owns everything else. Both files committed. Both read on every run.
Artifact 04
product-pulse
A scheduled task that pulls metrics, errors, conversions every morning into a markdown file. Without a pulse, you ship blind. With it, last week's drop shows up in tomorrow's planning. The system gets sharper every cycle.

The agent is a user of your design system

This is the move that separates agentic-native designers from "AI-augmented" ones. The augmented designer treats the agent as a tool. The agent-native designer treats it as a user, a collaborator, and a junior teammate, all at once. Document what the agent reads. What it can call. What it tends to get wrong. The agent persona sits next to your human personas. The components it consumes are documented in machine-legible form, not just visual mockups.

Once you make this shift, design system work stops being a chore and becomes the highest-leverage thing on your week. Every line you write about how a Button should behave teaches every future agent how every future Button should look. The work compounds across every screen the team will ever ship.

The augmented designer treats the agent as a tool. The agent-native designer treats it as a user, a collaborator, and a junior teammate, all at once.

How Designers Work · §08
09 · How We Work Together [9/11]

Smaller teams. Denser presence. Pulse is the meeting.

The big unlock is this: you're not handing off anymore. Product doesn't hand a PRD to engineering, who hands a build to QA, who hands a release to ops. The substrate holds the whole arc. The handoffs go away. The relays disappear. What's left is the actual work, plus the people doing it together.

Smaller pods
Two to four people plus their crews
A unit of value used to need ten people across five functions. Now it needs two to four people and the crews they spawn. Cross-functional from day one because the seats are fewer.
Denser presence
Real conversations, not status
Status is in Pulse. So when you're with a teammate, you're not catching up. You're actually working together — debating the spec, watching the same pulse, judging the same PR side by side.
Async by default
Plans, not meetings
A spec lives in the substrate. People comment. People sign off. Decisions don't require a calendar. When humans gather, it's because something needed humans gathered — vision, judgment, relationship.
Live, not retrospective
Meetings happen in front of the pulse
No one walks into a meeting and asks "where are we?" The room already knows. The meeting is for the questions the room can't answer — the human ones.
Three archetypes
Builders, owners, leaders
Builders make and run things. Owners own outcomes end-to-end. Leaders set example with the tools in their own hands. If you're not one of those three, you're middleware — and the substrate does middleware better than you can.
The room is bigger
Your teammates include agents
A pod of three people plus their crews can do work that used to need a department. Your circle of coworkers just got wider. Some of them will be people. Some of them won't. The covenant covers them all.

The most important meeting isn't on the calendar. It's the moment you and your crew look at the same pulse and ask, "is this the right thing built well?"

How We Work · §09
10 · The Maturity Ladder [10/11]

Six rungs of working with agents. We start at L3. Not because we're elite — because the substrate is.

Most teams climb this ladder one rung at a time, painfully. They write a CLAUDE.md after they get burned. They spawn a reviewer agent after the third missed bug. They trust an agent's plan after a year of line-by-line reviewing. You don't have to climb. You walk into a room where most of this is already built.

L0
Manual
Hand-typed code
Code written line by line. Research via docs. Debugging via print statements. Built great software for decades. Not fast enough for the work in front of us.
Not a real option for us. Not because manual coding is wrong — because we have a substrate that makes it unnecessary.
L1
Chat-based
AI as a smart reference
You ask Claude or ChatGPT for help. You copy answers back. AI is a reference tool. You're still doing all the work yourself.
The floor below which no Firma builder operates. If you're at L1, you're just getting started. Welcome.
L2
Agentic, line-by-line
Approve every step
The agent reads files and makes changes. You approve every action, line by line. This is the trap most teams die on. They never trust enough to step away.
A pass-through, not a destination. Don't get comfortable here. The plateau is real.
L3
Plan-first, PR-only
The Firma front door
You write the spec. The crew implements. You judge the PR. You step out of the line-level. You step into intent and judgment. The conscience in the floor catches what manual review used to.
This is where new Firma builders begin. Not climb to — begin at. The substrate is built so you can.
L4
Idea to PR
Describe the outcome
You describe the outcome. The crew researches, plans, builds, tests, reviews, ships a PR. Your involvement shrinks to idea, judgment, merge.
Where most builders should live day-to-day once they have a few weeks of feel for the substrate.
L5
Parallel fleet
Commanding crews from anywhere
Crews run in parallel in the cloud. You aren't tied to a laptop. You direct multiple workstreams from a phone, a beach, a coffee shop. You moved from individual contributor to commanding a fleet.
The destination. The Foundry pattern at full scale. This is what we're building toward, not from.

The reason we can start everyone at L3 is that we built the substrate first. The conscience in the floor, the crews, Pulse, the Foundry — they're already running before you arrive. You don't have to invent agentic-native engineering on top of a SaaS workflow. You walk in. You write a spec. You watch a crew. You judge a PR. You teach the system. That's the job from day one.

11 · Three Judgments & the Beliefs to Lay Down [11/11]

What to ask before merging. What to let go as you start.

Two short lists. The first is what you ask of every PR a crew brings you. The second is what you'll quietly need to put down — the engineering muscle memory that served you well before, and gets in the way now.

The Three Judgments

Every time a crew hands you something, ask three things. Out loud, of the agent, every time. They surface where the work is real and where it isn't:

Judgment 01
"What was the hardest decision you made here?"
Forces the crew to surface where the tricky parts were — where it had to make a judgment call. That's where you focus your review.
Judgment 02
"What alternatives did you reject, and why?"
Reveals the option space the crew considered. Catches if a clever path was missed — or if a wrong path was taken for a good-sounding reason.
Judgment 03
"What are you least confident about?"
Models know where they're shaky. You just have to ask. The least-confident answer is where your judgment is worth the most.

What to lay down

Some of what made you a great engineer before is going to feel like it's slipping. Some of it should slip. Here's what to put down, gently:

Lay down
"Code must be written by hand"
It must be good. It must solve the right problem. Whether human fingers or agent fingers typed it doesn't matter. The job is the result, not the typing.
Lay down
"Every line must be manually reviewed"
Manual review is one way to ensure quality. The substrate's reviewers, tests, and conscience are another. If you don't trust the system, fix the system — don't compensate by gatekeeping every line.
Lay down
"First attempts should be good"
First attempts are usually rough. Second attempts are still rough. The goal is iterating fast enough that the third attempt lands in less time than the first one used to. That's the system working.
Lay down
"More typing equals more learning"
Understanding matters more than muscle memory. You learn by reviewing, by catching mistakes, by knowing when the agent is wrong. The builder who reviews ten implementations sees more patterns than the one who hand-typed two.
Lay down
"Code is self-expression"
The code was never really yours. It belongs to the team, the product, the people it serves. Letting go is liberating. No attachment means feedback is easier, refactors are cleaner, and the work is the work.
Lay down
"Less typing means less work"
Directing a crew requires more thinking, not less. Fewer keystrokes. More decisions. More context-loading. More taste-encoding. The work didn't shrink. It moved up the stack.
Lay down
"Generating fifty options and picking one is design"
It isn't. That's curation. The agent did the work and you signed your name on it. Always run a critique pass — ask the agent to argue against its own output. Real design is in the spec and the judgment, not the picking.
Lay down
"Skipping the artifacts saves time"
It costs time. The forty minutes spent on a strategy file or a design file is the most leveraged forty minutes of the week. Skip the artifacts and the agent guesses what you mean — and you spend the same five days, with worse output.

Putting these down is uncomfortable. That's normal. The discomfort is the formation. What you'll find on the other side is that you were doing more typing than thinking, and the new shape lets you do more thinking than you've ever done before. That's the gift of the new posture, once you've walked into it.

Spec. Crew. Judge. Compound.
The next loop is sharper than this one.
◆ The Charge

This is the rare moment when you get to build the new thing from the first line. Don't waste it.

Most engineers in the world right now are trying to bolt agents onto a workflow that wasn't designed for them. They're paying for it in friction, in tooling sprawl, in code that gets harder to work with every quarter. You don't have to do that.

You're walking into a room where the floor was poured for this. THEOS is built for you and your agents to share. A heartbeat runs underneath; DNA Protocol listens to it and catches drift before it compounds. The conscience is in the substrate, so you can stay in flow on the work that's actually yours. Spec, crew, judge, compound — that's the rhythm. Tuesday will be sharper than Monday. Next month will be sharper than this one. Every loop will leave the substrate smarter than it found it.

And the work itself will feel different. Less keystroke-grinding. More thinking. More judgment. More vision. More of the things you got into engineering for in the first place.

We're not asking you to be a different person. We're asking you to lay down some old habits, pick up some new ones, and step into a posture where the agents are coworkers, the substrate is wired, and the bottleneck moved from your fingers to your taste. You'll be more leveraged than you've ever been. You'll ship more. You'll worry less about the small stuff. You'll spend more time doing the work that actually mattered to you.

There's no certification for this. There's no badge. There's only the things you've shipped, and the speed at which you ship the next one. Open the folder. Write the spec. Dispatch the crew. Judge the PR. Teach the substrate. Ship the URL. Then do it again, sharper.

You don't subscribe to THEOS. You live within it.

The agents are the operators. You are the conscience and the judge.

The substrate carries the load. You carry the vision.

Spec. Crew. Judge. Compound. Repeat.

Building Agentic-Native · A Vision Cast for the Team · Firma Labs · pairs with: agentic-first · theos-system-spec · dna-protocol