Come to Work AI Apprenticeship — Year One Studio

1. Arrival and AT lab: Week 1. Lock a reference VS Code profile, confirm assistive-technology setup, and bring up the classroom stack without a mouse.
   • AT profile interview and reference VS Code profile
   • Command-palette Olympics and six-item stack bring-up
2. Agent as collaborator, not oracle: Weeks 2–3. Teach bounded prompting, accessible diff reading, and approval discipline. Cline screenshot-browser loops are forbidden as a required path.
   • First bounded agent task with native HTML constraints
   • Diff-reading lab with a planted dangerous diff
   • Git as memory: intentional commits and secret hygiene
3. REAB and the product you are about to build: Weeks 4–5. Convert 'build Slack with AI' into requirements, expectations, alternatives, and behavioural expectations an employer can implement from.
   • Tear apart a bad brief and rewrite it as REAB
   • Accessibility acceptance-criteria workshop
   • Alternatives studio and AI behavioural expectations
4. Accessible interface patterns for conversation software: Weeks 6–7. Semantic message lists, labelled compose, live regions, pause control, and focus restore. The agent may scaffold CSS; the learner owns the accessibility tree.
   • Semantic message list and labelled compose field
   • Live-region lab: silent, polite, and assertive
   • Focus restore and pause control for incoming messages
5. SurrealDB as the product's memory: Weeks 7–8. CLI-first SurrealDB, schema, postgres-wire CRUD, record permissions whose passing test is a failed SELECT, live queries, and SurrealMCP.
   • Schema, seed, SurrealQL CRUD, and one postgres-wire path
   • Two-user permissions lab
   • Live queries wired to the pause control, plus SurrealMCP
6. AI writing, semantic search, harden, and present: Weeks 9–12. Permission-aware retrieval, disclosed AI compose, pair AT swap, portfolio, and capstone viva. Same competencies, rising evidence quality.
   • Embedding pipeline and permission-aware semantic search
   • AI compose with disclosure in UI and schema
   • Pair AT swap, portfolio assembly, and viva rehearsal

Duration

9 months (12-week classroom + 12-week employer team project + 12-week on-site)

• By the end of the classroom block, learners will operate a professional AI-assisted development environment entirely through keyboard and their own assistive technology, and will produce a supervision log that shows they approved, rejected, or revised agent actions with a stated rationale.
• By the end of the classroom block, learners will deliver a REAB product brief for the team-messaging workspace that an employer reviewer can act on, including testable accessibility acceptance criteria and behavioural expectations for the product, the user, and the AI features under uncertainty.
• By the end of the classroom block, learners will demonstrate a running accessible team-messaging prototype with channels, messages, simple permissions, pausable notifications, disclosed AI writing, and permission-aware semantic search, proven by a screen-reader or braille walkthrough of their own product.

• AT-optimized AI development workspace: Learner configures and operates an assistive-technology-optimised AI development environment entirely through keyboard and their own screen reader or braille display, producing durable, portable artifacts (VS Code Screen Reader Optimized profile, gitignored key handling, README, .env.example, AGENTS.md) that a second blind or low-vision apprentice can rebuild the stack from on unfamiliar hardware.
  Criteria: . Levels: [None]
• Human-in-the-loop agent supervision: Learner supervises a coding LLM (Cline as primary; CLI plus MCP fallback when the panel itself is inaccessible) as a human-in-the-loop collaborator: frames every task with a five-slot bounded prompt, marks every proposed edit or terminal command with an accept, reject, or revise decision plus a written rationale read from the Accessible Diff Viewer, and verifies the merged behaviour on a running system using their own assistive technology rather than the model's summary of its own work. Transfer-ready is scored only when this same rubric is re-applied against transfer-context evidence (a partner-issued ticket for Task framing, a teammate's shared repo for Approval discipline, and a partner UI at the A5 mid-phase checkpoint for Verification); within the A2 assessment window the achievable ceiling on this competency is Proficient.
  Criteria: . Levels: [None]
• REAB product specification: Learner produces docs/reab.md — a Requirements / Expectations / Alternatives / Behavioural-expectations product brief for the vertical-slice team-messaging capstone — that an employer reviewer can implement from without the apprentice present, including at least six testable accessibility acceptance criteria and behavioural rules for product, user, and AI-under-uncertainty.
  Criteria: . Levels: [None]
• Accessible realtime messaging interface: Learner constructs and defends an accessible realtime team-messaging vertical slice that a blind, low-vision, or Deafblind teammate can operate end-to-end from keyboard and their own AT, with an explicit announcement policy, focus-restore behaviour, permission-enforced persistence, and disclosed AI compose in both UI and ai_assist_event schema.
  Criteria: . Levels: [None]
• Permissioned multi-model data and retrieval: Learner implements a permissioned SurrealDB multi-model schema so record-level permissions, live updates, and permission-aware retrieval combine to guarantee that no search result, notification, or AI-tool query outranks the currently signed-in user, and proves that guarantee with a failed-SELECT-as-wrong-user transcript plus a permission-aware hit/miss/refuse trio in the A4 evidence bundle.
  Criteria: . Levels: [None]
• Professional demonstration and AI disclosure: Learner drives their own product end-to-end through their own assistive technology across the eight required keyboard paths, produces a partner-auditable evidence trail (supervision log with accept/reject rationales, at least one planted-dangerous-diff reject with an AT or scope reason, an ownership split of learner-owned / agent-drafted / jointly revised behaviour, and an ai_assist_event schema cross-check to the UI disclosure copy), and defends the work in a ten-minute employer-style viva against the five A6 challenge questions — with the same rubric scoring classroom, partner, and host placement blocks.
  Criteria: . Levels: [None]

• A1 Workspace practical
• A2 Agent supervision log and diff defense
• A3 REAB product brief
• A4 Messaging vertical slice
• A5 Accessibility verification protocol
• A6 Capstone viva and portfolio