README.md

Groundwork Curriculum

Part of Groundwork — the AI Readiness system by Adam Abdalla. Free. Ungated. MIT.

The free, ungated curriculum arm of Groundwork. A five-level path from AI novice to AI builder to AI contributor.

How this relates to the Groundwork site: the site is the readiness doctrine, six standing orders you build once for your business. This Curriculum is the capability path, five levels of skill you climb yourself. Different axes, one system. Full mapping: GROUNDWORK-AND-FLUENCY.md.

Two tracks at every level: individual and company. Two voices at every level: the standard docs and an ELI10 companion.

Start with the letter: shared/cover-letter/short-wake-up-letter.md — 5 minutes. Or the full case: shared/cover-letter/field-guide-manifesto.md — 30 minutes.

The five levels

See LEVEL-MAP.md for the visual map.

Where to start

• Start reading: shared/cover-letter/short-wake-up-letter.md (5 min)
• Get convinced: shared/cover-letter/field-guide-manifesto.md (30 min)
• Start doing: level-1-essentials/00-start-here/30-day-starter-sequence.md
• Set up your tools: shared/starter-packs/ — ready-to-paste custom instructions for Claude, ChatGPT, Perplexity, Claude Code, Cursor, Grok, Gemini, and the Anthropic API

Not sure this is the right program for you?

Read GROUNDWORK-AND-FLUENCY.md. The Groundwork Curriculum assumes your files, identity, and knowledge are already in a shape AI can use. If they aren't yet, the Groundwork program is the readiness prerequisite. Both are free to start; both are open. Neither gates its content.

The ELI10 layer

Every technical section in this program has a matching plain-language companion in the eli10/ folder at that level. Think of it as the same information explained to a curious 10-year-old — no jargon, real analogies, why-it-matters framing.

If a section feels dense, jump to its ELI10 twin first, then come back.

Program design principles

1. Two tracks, every level — individual + company
2. Two voices, every level — technical + ELI10
3. Every doc is opinionated — no placeholder shrug docs
4. Every workflow has an eval — measurable, not vibes
5. Every autonomy step is earned — HITL before full auto
6. Governance in front of scale — policy before pilots
7. Continuous refresh — the field moves; the kit moves with it

Ownership

You. Fork it. Extend it. Prune it. Ship your own kits from it.

License

MIT (see LICENSE) — use it, remix it, teach with it.

GROUNDWORK-AND-FLUENCY.md

Groundwork and the Curriculum: how the two programs fit together

TL;DR. Groundwork is the readiness program. The Groundwork Curriculum is the capability program: five levels that build AI fluency in you and your team. Groundwork gets your identity, knowledge, and rules into a shape AI can work on. The Curriculum teaches you to use AI well and, at the higher levels, to build the agents, tools, and evaluation systems that run on top of that foundation. They are sequential, not competing. Neither is gated. You can start in either program at any order or level that fits where you already are.

• Groundwork: groundwork.adamabdalla.com
• The Curriculum: this repo. Start at README.md.

One axis note before anything else, because it trips people up: Groundwork has six standing ORDERS. The Curriculum has five LEVELS. Orders are things you build once for your business. Levels are stages of skill you climb as a person or a team. The numbers are not the same axis and never map one-to-one.

Two programs, one arc

Groundwork's core thesis is "do not automate chaos." It is six standing orders built in sequence, producing fifteen plain-text documents, with one scoped AI Operator activated at order four:

• Orders 1 to 3, the foundation, required first: Order 1 Know me (Identity: who you are and how you are allowed to behave), Order 2 Know my information (Knowledge: what the system is allowed to trust and use, including the Context Vault and Authority Classification), Order 3 Know your limits (Governance: what the system is allowed to do).
• Order 4, activation: Help me operate. One scoped AI Operator, running under the rules the first three orders wrote.
• Orders 5 and 6, the standing half: Order 5 Remember (Continuity) and Order 6 Improve (measurement and the operating rhythm). From activation on, the system runs one loop: capture, understand, decide, act, verify, remember, improve.

Automation enters at order four, and only inside the three zones Order 3 defines: Auto-Execute, Draft & Wait, Never Automate. The default rule: unlisted work falls to Draft & Wait, never to Auto-Execute.

The Curriculum starts where a serious learner actually needs help: not "how do I organize my business" but "how do I get genuinely good at this, and eventually build systems without shipping something that embarrasses me." Level 1 assumes you can already talk to an AI product. Level 2 assumes you are ready to build. Level 3 and up assume you are operating AI systems for other people.

The overlap between the two is deliberate and small. Groundwork Order 4, the Operator, is the exact bridge: it hands off to Curriculum Level 2 the moment you want to build automation as software rather than as prompt-and-approval loops.

Full mapping: six orders to five levels

Groundwork's Readiness Scorecard names your weakest order. If it names Order 1, 2, or 3, run Groundwork first: the Curriculum assumes that foundation exists. If the foundation stands and your vault is real, Level 1 will feel like a natural next chapter. If your Operator is already running, you belong in Level 2.

Where the two programs agree, verbatim

Both programs teach the same three-zone approval model because it is the correct one:

In Groundwork these are the three zones of Order 3, Know your limits. In the Curriculum the same model appears as HITL propose-write defaults: the companion repos (belle-mcp-server, lease-abstractor, support-triage-agent, diligence-agent) all default to Zone 2 behavior, propose-write with auto_send off. That is on purpose. It is the same zone Groundwork tells you to default to when in doubt. Nothing in either program ships with autonomous send turned on.

Where the two programs deliberately differ

If you are a technical builder, the Curriculum is almost certainly what you need. If you are an owner who has never systematically organized the business's information, run Groundwork first: coming to the Curriculum without a vault and an approval framework in your head means you will build agents that automate a mess.

How to enter each program

Enter Groundwork if any of these are true: - Your files live in Downloads, Desktop, and email attachments - You explain the same preferences to AI every time you open a new chat - You have no rules for what AI is allowed to do without asking you - You worry about losing everything if your current AI provider changes terms - You want a Day One, a scorecard, and a finished system at the end

Enter the Curriculum if any of these are true: - You already have a Markdown notes system or the equivalent - You know what a system prompt is and have written one that matters - You want to build an agent, an MCP server, or a RAG pipeline this month - You want evals, not vibes - You are hiring or leading people who will build these things

Enter both, in sequence, if you are building AI capability for a real business: Groundwork gets the foundation right, the Curriculum builds the skill and the systems on top of it.

Cross-links in this repo

• Level 1 · Essentials README treats a knowledge base you own as the recommended starting condition; Groundwork Order 2, Know my information, is where that vault gets built
• Level 2 Company · 02-agent-governance references Groundwork Order 3, Know your limits, for the personal analogue of the org approval matrix, and Order 5, Remember, for access and estate rules
• Level 2 Company · 08-cross-team-workflows references the Groundwork zone model as the individual analogue of the org-wide approval matrix

No file in this repo requires Groundwork to make sense. You can jump into any level, any module, and get value. Groundwork is the recommended prerequisite for owners whose foundation is not yet set. It is not a paywall.

One-line summary for readers

Groundwork gets your information into a shape AI can use. The Curriculum teaches you to use it well and build systems on top of it. Run them in that order and both are cheaper and faster than doing this ad-hoc.

LEVEL-MAP.md

Level Map

The full path from novice to frontier. See shared/graphics/level-map.svg for the visual.

Level 1 — Essentials

Prereq: none. Exit criteria: - You have a master system prompt and About-Me block - You have a personal prompt library (20+ prompts) - You know which model to route which task to - You can pass the 15-question jargon fluency test - You've shipped at least 5 real work outputs using AI - For the company track: AUP signed, accounts procured, first two department copilots piloting

Unlocks: ability to build.

Level 2 — Intermediate

Prereq: Level 1 exit criteria met. Exit criteria: - You've built 3 real custom agents on your own stack - You've deployed at least one MCP server - You've built a real RAG pipeline over your own docs - You have an eval harness with regression testing - You understand cost engineering and have measured cost per workflow - For the company track: eval framework live, observability platform running, first agents in production

Unlocks: ability to ship AI to real users.

Level 3 — Advanced

Prereq: Level 2 exit criteria met. Exit criteria: - You've shipped an AI product or feature to real users, with SLOs met - You've orchestrated a multi-agent workflow in production - You've handled at least one real drift or incident - You've done a model bake-off with data-driven decision - You've hardened one workflow against adversarial input - For the company track: full observability, incident response tested, cross-department scaling underway

Unlocks: systems design.

Level 4 — Professional

Prereq: Level 3 exit criteria met. Exit criteria: - You've fine-tuned a model with measurable impact - You've led a red-team exercise - You understand training-stack internals (attention, MoE, quantization, inference) - You can teach any Level 1–3 concept - You can lead an AI team or an AI initiative - For the company track: applied research team stood up, safety program mature, cross-industry benchmarking

Unlocks: practitioner-grade work.

Level 5 — Frontier

Prereq: Level 4 exit criteria met. Exit criteria: - You've contributed original work: a paper, a novel system, a widely-used tool, or open-weight training - You engage with the research community - You have a point of view on where the field is going and act on it - For the company track: category-defining product, open-source contributions, industry-recognized capability

Unlocks: contribution to the field itself.

Movement between levels

• You don't have to be uniform. A person might be Level 3 on RAG, Level 2 on agents.
• A company might be Level 2 on governance, Level 1 on department deployments.
• The map is a compass, not a report card.

Timing

Realistic timelines assume 4–8 hours/week of dedicated effort alongside real work:

Faster is possible with full-time focus and prior software background.