Anusheel Bhushan

Timeline

1. 26 Oct 2026Day 40 · Week 8: Enterprise deployment, growth, research, release

   Shipped

   Deployed WorkBot live on Vercel with Neon, extended docs/architecture.md with deployment, operations and numbers, added decision D29, and fixed the release smoke suite's customer.

   • Architecture documentation
   • Product demonstration

   Final release

2. 26 Oct 2026Day 39 · Week 8: Enterprise deployment, growth, research, release

   Shipped

   Built a locked-DuckDB analytics layer with five versioned SQL metrics and data tests, an analyst agent that shows its SQL, and a published /research dossier.

   • Agentic research
   • BI with agents
   • Analytics

   Research and analytics with agents

3. 26 Oct 2026Day 38 · Week 8: Enterprise deployment, growth, research, release

   Shipped

   Built AEO site pages with structured data, robots, sitemap and llms.txt checked by check:aeo, a weekly citation measurement, account scoring, and compliant outreach with one-click unsubscribe.

   • Answer engine optimization
   • Growth automation
   • Outreach automation

   AEO and growth automation

4. 26 Oct 2026Day 37 · Week 8: Enterprise deployment, growth, research, release

   Shipped

   Added docs/research.md usability findings, release checklist and review page, Terraform uptime check and billing budget, and dead-letter document names in admin jobs.

   • User research
   • Discovery interviews
   • Usability testing
   • Customer cloud deployment
   • Privacy and personal data

   User research, and ship to a customer cloud

5. 26 Oct 2026Day 36 · Week 8: Enterprise deployment, growth, research, release

   Shipped

   Added a per-environment rotating KMS key with audit logs, per-tenant namespaces with network policies, a squid egress allowlist proxy, a load bench, and scaling fixes.

   • Infrastructure isolation
   • Key management
   • Tenant isolation
   • Scale thinking

   Isolation, keys and scale thinking

6. 26 Oct 2026Day 35 · Week 7: Documents, channels, process, cloud

   Shipped

   Added Terraform roots for GCP Cloud Run, Azure VM and Kubernetes, a Helm chart, a rollback release script, and GitHub Actions plan/approve/apply workflows.

   • Terraform
   • Infrastructure as code
   • Containers
   • Delivery pipelines

   Infrastructure as code: Terraform and deployment targets

7. 26 Oct 2026Day 34 · Week 7: Documents, channels, process, cloud

   Shipped

   Added a multi-stage Dockerfile, a compose.yaml running web, realtime, Postgres, self-hosted Inngest and a Playwright browser, plus platform-portable config for cloud deployment.

   • AWS
   • GCP
   • Azure
   • Neo-clouds
   • Cloud cost guardrails

   Cloud services: AWS, GCP, Azure and the neo-clouds

8. 26 Oct 2026Day 33 · Week 7: Documents, channels, process, cloud

   Shipped

   Added Postgres domain events with consumer positions and dead letters, per-owner concurrency fairness, 202 queued documents API, a Jobs admin page, a load test, and resumable SSE chat streams.

   • Queues
   • Pub/Sub
   • SSE and WebSockets
   • Serverless
   • Process lifecycle

   Process lifecycle: queues, pub/sub, SSE, WebSockets and serverless

9. 26 Oct 2026Day 32 · Week 7: Documents, channels, process, cloud

   Shipped

   Built a shared server-side conversation store, Twilio SMS/WhatsApp signed webhooks with status callbacks and a retrying outbox, and a Slack bot via the Chat SDK.

   • Multi-channel chat
   • Twilio
   • WhatsApp
   • iMessage
   • Slack
   • Vercel Chat SDK

   Multi-channel chat: web, SMS, WhatsApp, iMessage and Slack

10. 26 Oct 2026Day 31 · Week 7: Documents, channels, process, cloud

    Shipped

    Added byte-based format detection, PDF text-layer/scan routing, DOCX tracked-change reading, DuckDB SQL over XLSX sheets, a Sheets admin UI, and per-path format evals.

    • Document processing
    • PDF
    • DOCX
    • Spreadsheets
    • Large tables

    Documents and spreadsheets: PDF, DOCX, XLSX and large tables

11. 25 Oct 2026Day 30 · Week 6: Architecture, integrations, security

    Shipped

    Added Postgres row-level security per tenant, users/roles/hashed keys with sessions, role checks on every route and tool call, sealed GitHub tokens, and injection tests.

    • Agent security
    • RBAC
    • Secrets management
    • Prompt injection defenses
    • Tenant isolation

    Agent security: RBAC, keys and data separation

12. 25 Oct 2026Day 29 · Week 6: Architecture, integrations, security

    Shipped

    Added layered customer config with a second tenant (harbor-dental), an Ohm-grammar rule language with tests, model-proposed rules, an entity generator, and Turborepo checks.

    • Config-first design
    • Domain languages
    • BNF grammars
    • Meta-compilers
    • Monorepos

    Config-first design, grammars and meta-compilers

13. 25 Oct 2026Day 28 · Week 6: Architecture, integrations, security

    Shipped

    Added a JSON rule table with a rules engine, a unified ticket/contact schema keeping raw provider records, cursor sync plus signed GitHub webhook and nightly reconciliation, and a per-tenant cache.

    • Data sync
    • Unified schemas
    • Rule engines
    • Caching

    Integration techniques: sync, unified schemas, rules and caching

14. 25 Oct 2026Day 27 · Week 6: Architecture, integrations, security

    Shipped

    Added an MCP client, wrapped GitHub's MCP server in three idempotent ticket tools with dry runs, synced tickets to Postgres via Inngest, and documented route comparisons.

    • Composio
    • Nango
    • MCP
    • SaaS integrations
    • Tool surface design

    Integration platforms: Composio, Nango and MCP

15. 25 Oct 2026Day 26 · Week 6: Architecture, integrations, security

    Shipped

    Moved the customer into content/customer.json with a Zod parser, collapsed the analyst's two specialists into one agent after measured comparison, and wrote the audit, principles and decision record.

    • AI system architecture
    • Architecture principles
    • Architecture audits

    Architecture principles: what to unlearn

16. 25 Oct 2026Day 25 · Week 5: SDLC at scale, frameworks, providers, evals

    Shipped

    Built a unified eval suite for chat, extract and triage with variance-aware gating, judge calibration, reviewed synthetic cases, recorded baselines and an /admin/evals scoreboard.

    • Evals
    • LLM judges
    • Eval-driven development
    • Evals in CI
    • System prompting

    Evals as the development loop

17. 25 Oct 2026Day 24 · Week 5: SDLC at scale, frameworks, providers, evals

    Shipped

    Added a task-based routing table (src/lib/routing.ts) with fallback chains and OUTAGE rehearsal, a pnpm routes eval CLI scoring providers with cost and p95, and docs/routing.md.

    • Provider abstraction
    • Model routing
    • Fallbacks
    • Open-weight models

    2 providers, 1 codebase

18. 25 Oct 2026Day 23 · Week 5: SDLC at scale, frameworks, providers, evals

    Shipped

    Built the research agent on LangGraph.js, Claude Agent SDK and OpenAI Agents SDK with shared tools, a 5-topic benchmark CLI, and a written comparison.

    • LangGraph.js
    • Claude Agent SDK
    • OpenAI Agents SDK
    • Agent frameworks

    Agent frameworks: LangGraph.js, Claude Agent SDK, OpenAI Agents SDK

19. 25 Oct 2026Day 22 · Week 5: SDLC at scale, frameworks, providers, evals

    Shipped

    Built scripts/fleet.mjs to dispatch fifty specified tasks to worktree and cloud agents, with a brief-reviewer, batched merge queue with revert, and agent-produced fixes.

    • Agent fleets
    • Cloud agents
    • Test automation
    • PR review at scale

    Agent fleets: 50 concurrent agents, cloud agents and test automation

20. 25 Oct 2026Day 21 · Week 5: SDLC at scale, frameworks, providers, evals

    Shipped

    Configured Claude Code, Codex CLI and OpenCode on one shared policy (AGENTS.md, MCP, permissions, synced skills), added a brain/ directory, brief templates, a config-agreement test, and a guide.

    • Multi-CLI development
    • Context architecture
    • Working practices

    Multi-CLI development and context architecture

21. 25 Oct 2026Day 20 · Week 4: Build AI: production readiness

    Shipped

    Added docs/architecture.md describing WorkBot as built and docs/design/brief-a.md designing and pitching a support platform, plus cache-pricing and CI fixes.

    • AI system architecture
    • Product thinking
    • Technical pitching

    Architecture design and product thinking

22. 25 Oct 2026Day 19 · Week 4: Build AI: production readiness

    Shipped

    Added model-call tracing with cost and latency, versioned prompts with traffic splitting, a 12-case golden eval with LLM judge, rate limiting, and hourly spend/error alerts.

    • AI testing
    • Evals
    • Langfuse
    • Observability
    • Deployment

    Testing, monitoring and deployment

23. 25 Oct 2026Day 18 · Week 4: Build AI: production readiness

    Shipped

    Added an Inngest document-approval workflow with checkpointed steps and step.waitForEvent review, a cron reminders function, an admin inbox page, and docs.

    • Inngest
    • Durable workflows
    • Human-in-the-loop

    Workflow orchestration with Inngest

24. 25 Oct 2026Day 17 · Week 4: Build AI: production readiness

    Shipped

    Built src/lib/extract.ts with Zod schemas via Output.object, retry-with-reason up to 3 times, per-field confidence, 4 document schemas, document intake, and an 18-document eval.

    • Structured output
    • Zod
    • Data extraction

    Structured output and data extraction

25. 25 Oct 2026Day 16 · Week 4: Build AI: production readiness

    Shipped

    Added a browser voice line with listen/think/speak state machine and barge-in interruption via POST /api/voice, plus an ffmpeg frame-sampling video timeline pipeline.

    • Voice AI
    • Speech-to-text
    • Text-to-speech
    • Video understanding

    Voice and video AI

26. 25 Oct 2026Day 15 · Week 3: Build AI: agents and RAG

    Shipped

    Built a multi-agent research analyst in investigate.ts with knowledge-base and web specialists, browser retry fallback, per-agent logging and cost tracking, reports saved to tickets, and tests.

    • Multi-agent systems
    • Agentic RAG
    • Browser automation
    • Error recovery
    • Cost tracking

    The research analyst: orchestration, retrieval and the browser

27. 25 Oct 2026Day 14 · Week 3: Build AI: agents and RAG

    Shipped

    Added an AI scraper (src/lib/scrape.ts) extracting Zod-typed supplier records with rel=next pagination, and a Playwright screenshot-driven browser agent (src/lib/browser-agent.ts).

    • AI web scraping
    • Browser automation
    • Computer use

    AI scraping, browser use and desktop use

28. 25 Oct 2026Day 13 · Week 3: Build AI: agents and RAG

    Shipped

    Expanded the knowledge base to 52 heading-chunked documents with full-text and pgvector embeddings, added find/ls/cat/grep agent tools, and a top-5 versus agentic comparison CLI.

    • Agentic RAG
    • Embeddings
    • Retrieval

    Modern RAG: agentic retrieval

29. 25 Oct 2026Day 12 · Week 3: Build AI: agents and RAG

    Shipped

    Built a Claude Agent SDK explainer pipeline with researcher, writer and safety editor subagents, hook guardrails, revision handoffs and tracing, plus an intake triage route.

    • Multi-agent systems
    • Agent handoffs
    • Claude Agent SDK
    • Agent frameworks

    Multi-agent systems and frameworks

30. 25 Oct 2026Day 11 · Week 3: Build AI: agents and RAG

    Shipped

    Added a runAgent loop with step, dollar and time limits, an investigation agent using policy search and write_report, and a /investigate CLI command.

    • AI agents
    • Agent patterns
    • Agent design

    AI agent fundamentals

31. 25 Oct 2026Day 10 · Week 2: Build AI: foundations

    Shipped

    Added per-visitor long-term memory extraction stored in a memories table, a /settings page to forget facts, and token-budget context compaction with summarization.

    • Context management
    • Conversation memory
    • Long-term memory
    • Storage patterns

    Context and memory

32. 25 Oct 2026Day 9 · Week 2: Build AI: foundations

    Shipped

    Added a json-render catalog of 22 Zod-schema components, a streaming /api/generate route, and a /dashboards page that renders generated dashboards progressively.

    • Generative UI
    • Component catalogs
    • Progressive rendering

    Generative UI

33. 25 Oct 2026Day 8 · Week 2: Build AI: foundations

    Shipped

    Built a streaming web chat at /chat on the AI SDK via /api/chat, with inline tool call states, Markdown code blocks, browser-kept conversations and an admin assistant panel.

    • Next.js
    • Streaming UI
    • Chat app architecture
    • Tool use

    Full-stack AI chat app

34. 25 Oct 2026Day 7 · Week 2: Build AI: foundations

    Shipped

    Added four Zod-defined tools (calculator, page reader, Brave search, notes), a multi-round tool loop on Anthropic and OpenAI, and a cited /research chain.

    • Tool use
    • Multi-turn tool loops
    • Prompt chaining
    • OpenAI SDK

    Tool use and chaining calls

35. 25 Oct 2026Day 6 · Week 2: Build AI: foundations

    Shipped

    Added an Anthropic/OpenAI provider layer, WorkBot system prompt, a streaming `pnpm chat` CLI with model switching and /cost tracking, and live API tests.

    • Anthropic SDK
    • OpenAI SDK
    • Streaming
    • Prompt engineering
    • LLM cost control

    Anatomy of an AI app

36. 25 Oct 2026Day 5 · Week 1: Use AI

    Shipped

    Added read-only database and PostHog MCP servers registered in .mcp.json, moved requests to Postgres, added PostHog funnel events, and a DB-probing health check.

    • MCP
    • Custom MCP servers
    • Feature delivery

    MCP servers and complete feature development

37. 25 Oct 2026Day 4 · Week 1: Use AI

    Shipped

    Shipped several tagged features in parallel: request detail page, status changes, team/status filters, Markdown policies, a second site app, and folder-level CLAUDE.md files.

    • Git worktrees
    • Parallel agents
    • CLAUDE.md
    • Prompt engineering

    Multi-project and multi-feature workflow

38. 25 Oct 2026Day 3 · Week 1: Use AI

    Shipped

    Configured Linear, GitHub and Figma MCP servers in .mcp.json, wrote a 20-ticket backlog with acceptance criteria, and shipped tickets WB-5 through WB-8.

    • AI SDLC
    • MCP
    • Spec writing
    • Review-driven development

    The AI SDLC: feature development workflow

39. 25 Oct 2026Day 2 · Week 1: Use AI

    Shipped

    Added three Claude Code skills (/component, /api-route, /pre-deploy), a /review skill running bug, performance and security sub-agents in parallel, a Prettier hook, and Vitest.

    • Claude Code
    • Claude Code skills
    • Sub-agents
    • Hooks and slash commands

    Claude Code: skills and sub-agents

40. 25 Oct 2026Day 1 · Week 1: Use AI

    Shipped

    Scaffolded WorkBot with create-next-app (Next.js 16, TypeScript, Tailwind 4), added AGENTS.md and README, and set up shadcn/ui components.

    • Claude Code
    • AI dev environment
    • GitHub workflow
    • Vercel

    Dev environment and Claude Code fundamentals