Compendium · Compositions · Autonomy Ladders

Autonomy Ladders

Items

of 1002 total

Full-featured composed scenes — one per row so each design has room to breathe. Click for the live render, the spec, and the source.

Featured 14 Atoms 762 Compositions 240

№ 1321

Four-Rung Autonomy Hero

Scene STA go

Autonomy Ladder 4 rungs · 1 envelope

Four rungs of autonomy.
One governance envelope.

Each rung shifts the labour split between AI and human. @governor holds the envelope at every rung — that does not change.

L1 Rung 1 / 4

Augmented Assistance

Labour split

AI 20%Human 80%

Mode

AI suggests · @author decides

L2 Rung 2 / 4

Supervised Autonomy

Labour split

AI 50%Human 50%

Mode

AI drafts · @author edits

L3 Rung 3 / 4

Bounded Autonomy

Labour split

AI 75%Human 25%

Mode

AI ships · @author samples

L4 Rung 4 / 4

Strategic Autonomy

Labour split

AI 90%Human 10%

Mode

AI executes · @governor sets envelope

№ 1322

Augmented Assistance Detail Panel

Scene STA go

Rung 1

Autonomy Level

Augmented Assistance

AI / Human 20 / 80

AI does

The labour

·Drafts code, copy, queries
·Surfaces options & trade-offs
·Flags spec inconsistencies
·Generates test scaffolds

@author does

The judgment

·Reviews every diff before merge
·Owns intent + acceptance criteria
·Decides which suggestion lands
·Carries production accountability

@governor demands

Evidence

·100% pre-merge review
·Spec ↔ change traceability
·Reviewer signature on PR
·Reverts attributed to @author

№ 1323

Supervised Autonomy Detail Panel

Scene STA go

Rung 2

Autonomy Level

Supervised Autonomy

AI / Human 50 / 50

Shift from L1 → @author becomes editor, not co-author. Review shape changes line → outline.

AI does

The drafting

·Drafts whole modules / features
·Proposes implementation paths
·Writes the first test pass
·Self-applies house style

@author does

The editing

·Edits outline + structure
·Owns spec + acceptance criteria
·Approves every PR pre-merge
·Carries on-call accountability

@governor demands

Evidence

·100% pre-merge sampling
·Spec linked from every PR
·Diff annotated by @author
·Risk note for size > 200 LOC

№ 1324

Bounded Autonomy Detail Panel

Scene STA go

Rung 3

Autonomy Level

Bounded Autonomy

AI / Human 75 / 25

Shift from L2 → Review moves pre-merge → post-hoc. Rails become machine-checkable, not human-checkable.

AI does

The shipping

·Ships changes inside the rails
·Self-rolls-back on rail breach
·Emits structured telemetry
·Calls only typed actions

@author does

The sampling

·Defines the rails (spec + policy)
·Samples 10-20% post-hoc
·Tunes thresholds on drift signals
·Reviews the rail set monthly

@gatekeeper enforces

The rails

·Typed action surface only
·Policy gate on every ship
·Audit log + replay capability
·Auto-rollback on SLO breach

№ 1325

Strategic Autonomy Detail Panel

Scene STA go

Rung 4

Autonomy Level

Strategic Autonomy

AI / Human 90 / 10

Kill-switch armed

Shift from L3 → Rails (what to do) → envelope (what NOT to do). Sampling → audit.

AI does

The execution

·Pursues goals across sprints
·Plans + executes multi-step work
·Self-monitors against envelope
·Escalates on confidence < threshold

@governor does

The envelope

·Sets envelope (what may NOT be done)
·Reviews escalations & exceptions
·Audits the audit trail weekly
·Owns the off-switch

Kill-switch

The brake

·Always armed, never debated
·Single-actor authority (@governor)
·Snapshots state before halt
·Triggered by SLO, policy, or hunch

№ 1326

Autonomy Transition Flow

Scene STA go

Transition Flow 3 triggers · 1 governor

Rung transitions are events,
not drifts.

Teams move between rungs on discrete triggers. @governor is the only signature that commits the change.

Trigger 1

Incident

SLO breach or policy violation in last calibration window

Trigger 2

Baseline

Spec, policy or risk appetite has materially shifted

Trigger 3

Cadence

Quarterly calibration window opened — review due

Decision

@governor reviews evidence

Promote · hold · demote

Promote ↑ Climb a rung

·Sampling rate → lowered
·Rail set → widened
·New rung announced + dated
·AAR scheduled at +30 days

Demote ↓ Drop a rung

·Sampling rate → 100%
·Rails → tightened or frozen
·Rung step-down dated immediately
·Root-cause GTR opened

№ 1327

Hat Shift Across Autonomy Levels

Scene STA go

Hat × Rung 4 hats · 4 rungs · weight shifts

Hats do not vanish. Their weight redistributes.

Hat \\ Rung	L1 Augmented	L2 Supervised	L3 Bounded	L4 Strategic
Intent Architect	Med	Heavy	Heavy	Heavy
Composition Lead	Heavy	Heavy	Med	Light
Flow Steward	Light	Med	Heavy	Med
Governor	Light	Med	Heavy	Heavy

№ 1328

Sampling Strategy vs Autonomy

Scene STA go

Sampling Rate 100% → 5% across rungs

The quietest dial on the ladder.

What fraction of work do humans actually inspect at this rung? Honest answer or no climb.

100%

20%

Augmented

Pre-merge

Supervised

Pre-merge

Bounded

Post-hoc

Strategic

Audit

№ 1329

Risk vs Autonomy Matrix

Scene STA go

Risk × Autonomy 4 × 4 matrix

Not every job belongs at every rung.

Risk class \\ Rung	L1 Aug.	L2 Sup.	L3 Bnd.	L4 Str.
Low / reversible	OK	OK	OK	OK
Medium / observable	OK	OK	Sample	Sample
High / customer-facing	OK	Sample	Sample	Forbid
Catastrophic / financial	Sample	Forbid	Forbid	Forbid

№ 1330

Practices Per Autonomy Level

Scene STA go

Practices Per Rung Self-audit checklist

A rung is what you can prove.

L1 · Augmented 4 req · 1 star

REQSpec exists for every change
REQPre-merge review on 100% of PRs
REQSuggestion provenance recorded
REQ@author signs every merge
STARSuggestion-quality dashboard

L2 · Supervised 4 req · 1 star

REQL1 practices + outline-edit discipline
REQPR size cap (e.g. 400 LOC)
REQSpec ↔ PR link enforced in CI
REQReviewer rotation, no single-eye merges
STARRisk-tagged PR template

L3 · Bounded 5 req · 1 star

REQTyped action surface; no shell-out
REQPolicy gate in deploy pipeline
REQAudit log + replay capability
REQSampling cadence published
REQAuto-rollback on SLO breach
STARDrift detector on rail breaches

L4 · Strategic 5 req · 1 star

REQEnvelope spec — what the system may NOT do
REQKill-switch wired, drilled quarterly
REQWeekly @governor envelope audit
REQGoal ↔ metric divergence detector
REQException queue + SLA
STARIndependent red-team cadence

№ 1331

Governance Intervention Frequency

Scene STA go

Governor Interventions Per 100 changes · by rung

The intervention curve is not monotonic.

Climbing the ladder is not "less governance" — it is differently-shaped governance.

human-in-loop

edit-discipline strain

rails absorb load

envelope-edge cases

№ 1332

Kill-Switch Authority by Level

Scene STA go

Halt Authority Unbroken across rungs

The cord is always there.

Scope grows. Latency target tightens. Authority does not transfer to the system at any rung.

HALT · L1 ●

Scope

Single PR

Latency target

< 1h

Who pulls

@author

HALT · L2 ●

Scope

PR + branch

Latency target

< 30m

Who pulls

@author

HALT · L3 ●

Scope

Deploy + flag

Latency target

< 5m

Who pulls

@gatekeeper

HALT · L4 ●

Scope

Fleet + agent

Latency target

< 60s

Who pulls

@governor

Continuous authority → L1 — L2 — L3 — L4. One cord. One owner. No exceptions.

№ 1333

Autonomy Progression Timeline

Scene STA go

Progression 6 quarters · 1 team

A healthy climb has demotions in it.

Q1 L1

Spec discipline established
→

Hold
Q2 L2

Calibration steady
↑

Promote
Q3 L2

Rails + policy gates landed
↓

Demote
Q4 L3

Incident: policy gate bypass
↑

Promote
Q5 L2

Rail set hardened + drilled
Q6 L3

Now-state · 6 quarters in

№ 1334

Confidence Threshold by Level

Scene STA go

Escalation Threshold Confidence floor by rung

Less feedback in the loop ⇒ higher escalation floor.

L1 Augmented

L2 Supervised

L3 Bounded

L4 Strategic

№ 1335

Hat Conflict Mitigation Across Levels

Scene STA go

Hat Conflict Mitigation by rung

Higher rung. Stricter separation.

L1 · Augmented

Hat conflict policy

Allowed combos

Intent + Composition
Composition + Flow

Forbidden combos

Mitigation

Same person may wear multiple hats. Spec review covers the gap.

L2 · Supervised

Hat conflict policy

Allowed combos

Intent + Composition

Forbidden combos

Composition + Governance

Mitigation

Reviewer rotation. Composition Lead never approves their own PR.

L3 · Bounded

Hat conflict policy

Allowed combos

Intent + Flow

Forbidden combos

Composition + Governance
Intent + Governance

Mitigation

Independent @gatekeeper reviews rail-set + policy changes.

L4 · Strategic

Hat conflict policy

Allowed combos

Forbidden combos

Any pair with Governance

Mitigation

Hat separation enforced. @governor cannot also author the envelope they audit.

№ 1336

Autonomy and Accountability Map

Scene STA go

Accountability Map 5 rows · 4 rungs

Autonomy moves. Accountability does not.

The harm row stays on @governor at every rung. That is the invariant of the ladder.

Accountable for \\ Rung	L1 Aug.	L2 Sup.	L3 Bnd.	L4 Str.
Intent (the why)	@author	@author	@author	@governor
Code / change (the how)	@author	@author	AI	AI
Deploy decision	@author	@author	@gatekeeper	@gatekeeper
Outcome in production	@author	@author	@author	@governor
Harm to users / business	@governor	@governor	@governor	@governor ↑ invariant

— Full detail →

—

// source

Alpha · Honest about it

The framework is real.
The community is forming now.

DoCoDeGo is in Alpha. The framework is documented, the practices are battle-tested at small scale, and the next release is being shaped in public.

If it produces anything, it should produce engineers and teams who think more clearly about what they are building and why.

Two doors in

Join the Discord

Discord is where specs are debated, the framework gets sharper, and decisions land in writing. The conversation is the artefact.

Autonomy Ladders

Augmented Assistance

Supervised Autonomy

Bounded Autonomy

Strategic Autonomy

The framework is real.The community is forming now.

The framework is real.
The community is forming now.