The Leverage Map
How we got here: the other maps
Ten other people and teams drew a ladder for how work moves to AI. Here they are, and where ours fits.
Most of them measure one thing: how much the AI carries. The Leverage Map measures two, what you can do and how much AI carries, and splits that into six small-business jobs, for people without a dev team.
Every entry is credited and links to its source
00The pattern
Most of them draw a ladder
Oldest first. One column per framework, one rung per level, all on the same scale, so the count shows at a glance. On a ladder the bottom rung is where the person does the work and the top rung is where the AI carries most of it. Pillars stand side by side, and a spectrum has no steps. Ours is last, drawn as a grid, because it has two axes.
- 0 no driving automati…
- 1 driver support
- 2 driver support
- 3 automated driving
- 4 automated driving
- 5 full driving automa…
01SAE Levels of Driving Automation (J3016)
SAE International · 2014
- No AI
- Emerging
- Competent
- Expert
- Virtuoso
- Superhuman
performance
- No AI
- AI as a Tool
- Consultant
- Collaborator
- Expert
- Agent
autonomy
02Levels of AGI for Operationalizing Progress on the Path to AGI
Google DeepMind · 2023
- the augmented LLM
- workflows
- autonomous agents
Anthropic · 2024
- operator
- collaborator
- consultant
- approver
- observer
04Levels of Autonomy for AI Agents
University of Washington; Knight First Amendment Institute · 2025
- frontier model capabi…
- verification
- context management
Replit (AI Engineer CODE 2025 talk) · 2025
- near-zero AI
- IDE agent with permis…
- IDE agent in YOLO mode
- wide IDE agent
- single CLI agent
- 3 to 5 parallel CLI a…
- 10+ hand-managed agen…
- building your own orc…
06The 8 Stages of Dev Evolution to AI (Welcome to Gas Town)
Independent (Gas Town) · 2026
- spicy autocomplete
- coding intern
- junior developer
- developer as reviewer
- engineering team led…
- dark software factory
07The Five Levels: from Spicy Autocomplete to the Dark Factory
danshapiro.com · 2026
- Chatbot
- Copilot
- Agent
- Autopilot
- Workflows
- Assistant
- Multi-agent
- Orchestrator
08The Eight Levels of AI Adoption
Every · 2026
- Unengaged
- Assisted
- Conversational
- Directed
- Parallel
- Autonomous
09AI Maturity Model (Building an Autonomous Engineering Org)
Block; Agentic AI Foundation (AI Engineer talk) · 2026
- Gated
- Assisted
- Parallel
- Supervised autonomy
- AI-native
Anthropic (Claude Code) · 2026
Two axes. Across, L0 to L5: what you can do. Up, A0 to A5: how much of the job AI carries. Six lines, one per job, each with its own star.
11The Leverage Map
San Diego AI Studio · 2026
01The ten
Each one, who drew it, and how it relates to ours
The shape is the source's own set of levels, in its own words. The line marked Ours says where it meets the Leverage Map. The axis tag says what the ladder measures.
- Read the source
SAE Levels of Driving Automation (J3016)
SAE International (with ISO, 2021 revision) · SAE International · 2014 (current revision 2021-04)
6 levels: Level 0 no driving automation through Level 5 full driving automation; Levels 1 and 2 are driver support, 3 to 5 are automated driving.
OursThis is the pattern most AI ladders borrow, including Shapiro's; our Delegation Altitude A0 to A5 rhymes with it on purpose, and the Operator Level axis adds what the driver can do.
- Read the source
Levels of AGI for Operationalizing Progress on the Path to AGI
Meredith Ringel Morris, Jascha Sohl-Dickstein, Noah Fiedel, Shane Legg and co-authors · Google DeepMind · 2023-11-02
Two separate scales: performance Levels 0 to 5 (No AI, Emerging, Competent, Expert, Virtuoso, Superhuman) and autonomy Levels 0 to 5 (No AI, AI as a Tool, Consultant, Collaborator, Expert, Agent).
OursIt is the closest in structure to ours because it keeps what the system can do apart from how much autonomy it is given; it rates AI models, while the Leverage Map rates a person's work.
- Read the source
Building effective agents
Erik Schluntz and Barry Zhang · Anthropic · 2024-12-19
A spectrum, not numbered levels: the augmented LLM, then workflows (prompt chaining, routing, parallelization, orchestrator-workers, evaluator-optimizer), then autonomous agents.
OursIt describes the same move toward the AI carrying more, from the builder's side, with the advice to use the simplest option that works, which is how our map treats each next altitude: earned, not assumed.
- Read the source
Levels of Autonomy for AI Agents
K. J. Kevin Feng, David W. McDonald, Amy X. Zhang · University of Washington; Knight First Amendment Institute · 2025-06
5 levels named by the user's role as the agent gains autonomy: operator, collaborator, consultant, approver, observer.
OursIt names each level by the role the human keeps as the agent carries more, which maps onto our Delegation Altitude, and it treats autonomy as a design choice separate from capability, the same split as our two axes.
- Read the source
The 3 Pillars of Autonomy
Michele Catasta · Replit (AI Engineer CODE 2025 talk) · 2025-12-22
3 pillars that let an agent run without technical supervision: frontier model capability, verification, context management; framed as supervised versus full autonomy.
OursHe builds for nontechnical users who cannot review code, the same people our map serves, and names verification as what lets the AI carry more without the person watching every step.
- Read the source
The 8 Stages of Dev Evolution to AI (Welcome to Gas Town)
Steve Yegge · Independent (Gas Town) · 2026-01-01
8 stages: near-zero AI, IDE agent with permissions, IDE agent in YOLO mode, wide IDE agent, single CLI agent, 3 to 5 parallel CLI agents, 10+ hand-managed agents, building your own orchestrator.
OursBoth Every's and Angie Jones's ladders were rebuilt from this one; it defines stages by developer tooling, while the Leverage Map defines levels by behavior across six business jobs so a non-engineer can place themselves.
- Read the source
The Five Levels: from Spicy Autocomplete to the Dark Factory
Dan Shapiro · danshapiro.com · 2026-01-23
6 levels (0 to 5) modeled on driving automation: spicy autocomplete, coding intern, junior developer, developer as reviewer, engineering team led by a PM, dark software factory.
OursEach level pairs how much the AI carries with the role the human now plays, which is the pairing our two axes make explicit, but his is scoped to software teams.
- Read the source
The Eight Levels of AI Adoption
Mike Taylor and Laura Entis · Every · 2026-06-02
8 levels: Chatbot, Copilot, Agent, Autopilot, Workflows, Assistant, Multi-agent, Orchestrator; each level delegates more work and more trust to the AI.
OursIt folds what the person does and how much the AI carries into one ladder; the Leverage Map splits those into two axes, and Every's own note that most knowledge workers sit at Levels 1 to 4 describes the people our map is built for.
- Read the source
AI Maturity Model (Building an Autonomous Engineering Org)
Angie Jones · Block; Agentic AI Foundation (AI Engineer talk) · 2026-06-28
6 stages (0 to 5): Unengaged, Assisted, Conversational, Directed, Parallel, Autonomous; it measures the engineer's relationship with agents, from no AI to delegating complete tasks that ship.
OursShe scores each engineer on how far they delegate to agents, which is close to reading our Delegation Altitude per person, and her champions program makes the same bet as our Kit and Sprint: a few trained operators lift everyone else.
- Read the source
Steps of AI Adoption
Boris Cherny · Anthropic (Claude Code) · 2026-07-16
5 steps (0 to 4) keyed to agent count: Gated (0), Assisted (~1), Parallel (~10), Supervised autonomy (~100), AI-native (1,000+); each step lists role, bottleneck, products and guardrails.
OursIt tracks one axis, how much work agents carry across an engineering org, which lines up with our Delegation Altitude, and it makes the same claim we do: tokens alone do not move you up, loops and guardrails do.
02Where ours fits
Two axes, six jobs, and a star on each line.
They reach the same ladder from different starting points, which is one signal it holds up. Ours was built from how we work with the businesses around us, then checked against these.
The distance between your two numbers is what a job still costs you in your own hours. That gap is what the map shows, and what the rest of the studio helps close.
- What the others measure
- How much the AI carries, mostly for engineering teams: agent counts, autonomy levels, stages of tooling.
- What ours keeps apart
- Level, L0 to L5, is what you can do. Altitude, A0 to A5, is how much of a job AI carries. Every stop carries both.
- Who it is for
- Small business owners, on six real jobs: Ops, Sales, Content, Answers, Media, Build. You place yourself in about a minute, with no dev team.