The lab, in full

What AR Labs has built and is building: voice agents, VOX OS and the AI agent orchestration tools behind it, apps and open-source Claude Code tools.

1. 01.01Salesbot: a Pipecat voice agent for sales discoveryVoice agentLiveAR Labs' Salesbot is a browser voice agent built on Pipecat that runs a short sales discovery conversation, reads answers back and closes with a summary.
2. 01.02Discovery Copilot for live sales callsConversation copilotBuildingAR Labs' Discovery Copilot is a prototype that follows a live sales call, ticks a discovery checklist with quoted evidence and suggests the next question.
3. 01.03VOX PresenterVoice sessionsBuildingAR Labs' VOX Presenter runs voice sessions about documents: a fast voice agent discusses a document aloud while a second agent writes analysis on screen.
4. 01.04Voice UIVoice interface kitPausedAR Labs' generic browser voice interface and Pipecat backend, taken from the first Salesbot as the start of a reusable voice layer for AI agents. Paused.
5. 01.05VOX FacilitatorFacilitation designDesigningAR Labs' design for a voice facilitator that reads a folder of project documents, then runs a spoken session from understanding to decisions to tasks.
6. 01.06VOX SchedulerCalendar assistantPausedAR Labs' design for an AI voice calendar assistant for Google Calendar: early prototypes and a documented web design. Paused at the design stage.
7. 01.07CoherenceVoice coaching appPausedAR Labs' Android voice-journaling and goal-coaching app: a spoken daily check-in with an AI guide, built as a staged conversation whose state lives in code.
8. 01.08VOX OS: personal AI operating system architectureAgent orchestrationDesigningVOX OS is AR Labs' design for an operating layer over AI agent work: one task graph, with terminal and browser views for one person steering many agents.
9. 01.09Vox Canvas: a browser task graph viewerTask graph viewerBuildingVox Canvas is AR Labs' keyboard-driven browser canvas for its task graph: a tree to move through, a panel for each task, and Taskgraph's status map in place.
10. 01.10VOX Torch: a terminal view over AI agent workTerminal interfaceBuildingVOX Torch is AR Labs' terminal interface over its live task graph: portfolio, line of work, single task, what waits on a person, and the agents running.
11. 01.11gc-torch: a terminal view for Gas CityTerminal observabilityBuildinggc-torch is AR Labs' read-only terminal view over Gas City, Steve Yegge's agent orchestrator: sessions, work, what waits on a person and an event feed.
12. 01.12Taskgraph: a status map for coding agent workWork status modelBuildingTaskgraph is AR Labs' command-line tool that reads a branch of the task graph with its git and pipeline state and maps where the work stands.
13. 01.13Goal Map / Agent Orchestrator: AI agent sessions around the month's goalAgent fleet viewBuildingAR Labs' Goal Map / Agent Orchestrator places every project on a spiral around the month's goal, with the AI agent sessions working on each in orbit.
14. 01.14AI software factory: coding agents with gatesAgent coding pipelineDesigningAR Labs' design for an AI software factory: coding agents take tasks from the task graph to merged code through gates, while a person orders and merges.
15. 01.15VOX Memory: research into memory for AI agentsAgent memory researchPausedAR Labs' research into long-term memory for AI agents: a framework evaluation, a rejected conversation-memory trial and a paused knowledge-graph design.
16. 01.16AI agents at AR Labs: the agents we run and what each doesAgent fleetBuildingThe AI agents AR Labs runs: a strategy agent, a marketing manager agent, a chat identity for laptop sessions, and coding, research and voice agents.
17. 01.17AI strategy agent: planning reviews from the year down to the dayStrategy agentBuildingAR Labs' AI strategy agent runs the lab's planning reviews, yearly to daily, with a person who decides, then records decisions and passes direction down.
18. 01.18AI marketing manager agent: Night Shift, on the Hermes runtimeMarketing agentBuildingAR Labs' AI marketing manager agent, Night Shift: it researches, recommends, briefs and measures on a server, while a person writes and approves.
19. 01.19Vox Prime: a Slack identity that lets coding sessions message AI agentsAgent identityBuildingVox Prime is AR Labs' identity in its shared Slack channel for laptop coding sessions, so a session can ask an AI agent for something and read the reply.
20. 01.20Resident AI agents: a multi-agent system with rolesMulti-agent systemBuildingHow AR Labs runs resident AI agents with their own roles, among them a marketing manager and a strategy agent, and how they work with coding agents.
21. 01.21AR Labs open-source releases and public repositoriesIndexAR Labs' public code on GitHub: what each repository holds, its state, and the licence GitHub reports for it, including where there is none.
22. 01.22VoxPlanAndroid appBuiltAR Labs' open-source Android app for hierarchical goal planning and time tracking, with daily time quotas and a gamified focus timer.
23. 01.23Vox ManifestorVoice appPausedAR Labs' voice-first Android app in which an AI 'genie' helps a person name a few wishes and map where they are now against where they want to be.
24. 01.24TaskGlowTimerWeb appBuiltAR Labs' keyboard-driven web focus timer that links each task to a goal, rewards every 30 minutes of work, and was the test bed for our design system.
25. 01.25Break It DownClaude Code commandBuiltAR Labs' open-source Claude Code slash command that turns a long, many-question AI response into a step-by-step conversation, one chunk at a time.
26. 01.26Coding Planner: a spec-driven workflow for Claude CodeCoding pipelineBuildingAR Labs' tooling for its coding pipeline in Claude Code: it routes coding work, plans it before any code is written, and checks each step with scripts.
27. 01.27Code health skill for Claude CodeClaude Code skillPausedAR Labs' prototype Claude Code skill for multi-pass code health reviews, based on Steve Yegge's advice on keeping agent-written code healthy.
28. 01.28Frontend DesignerDesign toolingPausedAR Labs' set of Claude Code agents that analyse a live website and extract its visual design into a reviewed prototype and reusable design tokens.
29. 01.29Second Brain Workshop: build a second brain with Claude CodeWorkshopPausedAR Labs' hands-on workshop series teaching coaches, freelancers and small-business owners to build a second brain with Claude Code from plain markdown files.
30. 01.30Cognitive Ecology (book)Book manuscriptBuildingThe book manuscript behind AR Labs' cognitive-science approach: Richard M. Thompson on attention, the unconscious, framing and goals, now being restructured.
31. 01.31Claude Certified ArchitectCertificationLiveRichard M. Thompson, who founded AR Labs, holds Anthropic's Claude Certified Architect – Foundations credential. We built a study tracker for the exam.
32. 01.32Bpai Dûay Gan (Go Together)Hackathon appBuiltA hackathon app co-built by AR Labs that pairs Thai locals with nomads in Chiang Mai for events and skill swaps, with chat translated by Claude.
33. 01.33arlabs.techWebsiteLiveAR Labs' public website: a navigable map of the lab as its front page, a reading layer for writing and project pages, and live tools such as a voice agent.

The vision and the ideas behind AR Labs' work with AI agents, voice interfaces and context engineering, explained for a newcomer.

1. 02.01The AR Labs visionVisionThe AR Labs vision for AI: voice in, concepts out, on an architecture that fits the limits of people and language models, held by individuals and small teams.
2. 02.02Mission and valuesVisionAR Labs builds with machine intelligence for four values: connection and collaboration, education and progression, quality of life for all, sustainability.
3. 02.03What we believe about AIBeliefAR Labs' positions on AI: conversation as a main interface, AI that augments people, people keeping judgement, speed with direction, governed agents.
4. 02.04The future of work with AI agentsBeliefAR Labs' outlook on the future of work with AI agents: agents carrying defined roles while people frame, supervise and judge, and what that asks of people.
5. 02.05An ecology of mind and machineBeliefThe idea behind AR Labs' tagline: people, AI agents and shared records form one system, each mind doing what it is good at, actions meeting a wider check.
6. 02.06The Loop: how AR Labs worksProcessThe Loop is the five-stage cycle by which AR Labs learns, experiments, produces, applies and shares, so that each piece of work feeds the next.
7. 02.07The AI interface architectBeliefThe practitioner AR Labs thinks AI needs: a designer of people-to-agent interfaces, combining agent engineering, voice, visualisation and production skill.
8. 02.08The lab aestheticBeliefAR Labs' visual principles: retro-futurist and vintage-computing inspired, readable first, calm for long thinking, people and AI drawn as collaborators.
9. 02.09VOX: voice in, concepts outConceptVOX, AR Labs' central idea: speak to AI agents in ordinary conversation and get back a visual map of the ideas and the work, sized for people and models.
10. 02.10Cognitive ecology for people and AI agentsFrameworkCognitive ecology, AR Labs' founding framework: small conscious attention frames a large unconscious process, and actions are judged by wider consequences.
11. 02.11Context engineering for AI agentsConceptContext engineering for AI agents as AR Labs practises it: choosing which instructions, knowledge, tools and memory enter a model's context window.
12. 02.12The harness paradoxConceptAR Labs' harness paradox: tightly constrained AI agents do more useful work than loosely configured ones, because each constraint is information.
13. 02.13Structured conversation for voice agentsConceptAR Labs' structured conversation pattern for voice agents: the language model talks freely while code tracks what has been collected and when to move on.
14. 02.14Logical levels in agent designFrameworkGregory Bateson's logical levels as AR Labs applies them to AI agent instructions: nested frames that each govern the level below, records kept outside.
15. 02.15The temperature of specificationConceptAR Labs' principle for planning AI coding agent work: start in design-level language and grow specific stage by stage, leaving code detail to later steps.
16. 02.16The intelligence is in the orderingConceptAR Labs' view of AI agent orchestration: every actor, human or agent, needs a task list in the right order, so orchestration is ordering and routing.
17. 02.17The agentic organisationFrameworkAR Labs' design for a company whose roles AI agents can fill: durable threads of work, seats agents occupy, handoffs, status up and direction down.
18. 02.18The second brainConceptAR Labs' approach to an AI second brain: your goals, history and preferences in plain files that an AI agent such as Claude Code reads each session.
19. 02.19Cognitive chunking for AI outputConceptAR Labs' method for presenting AI responses one piece at a time, in chunks sized to working memory, so a long answer becomes a conversation.
20. 02.20The acceleration paradoxConceptAR Labs' view of AI productivity: feeling faster is not being faster, and some work, learning above all, should not be sped up.

How AR Labs builds software with AI coding agents: spec-driven, test-first work, checks at every step, independent review and a person at the merge.

1. 03.01How AR Labs engineers software with AI coding agentsOverviewHow AR Labs builds software with AI coding agents: design first, a spec and tests before code, scripted checks, cross-model review and a person at the merge.
2. 03.02Software architecture with AI coding agentsOverviewHow AR Labs designs software that AI coding agents build: design before spec, deep modules, C4 diagrams, decision records, risk spikes and tracer-bullet slices.
3. 03.03Agentic engineeringOverviewWhat AR Labs means by agentic engineering: building software with and around AI agents by designing harness, context, tools, gates, verification and cost.
4. 03.04Design principles for AI-generated codeOverviewDesign principles AR Labs applies to code from AI coding agents: build only what was asked, deep modules, two cases before a pattern, one source of truth.
5. 03.05The coding pipeline for AI coding agentsFrameworkHow AR Labs takes coding work from a rough idea to reviewed, merged code with AI coding agents, a task graph and scripted checks at every step.
6. 03.06Spec-driven development with AI coding agentsPracticeHow AR Labs runs spec-driven development with AI coding agents: a campaign plan (PRP) and a per-task spec grounded in today's code.
7. 03.07Test-driven development with AI coding agentsRuleHow AR Labs' AI coding agents work test-first: a failing test, the least code to pass it, tests at deep-module boundaries and a ten-minute regression gate.
8. 03.08Evidence of completion for AI coding agentsRuleHow AR Labs gets evidence of completion from AI coding agents: a committed test, check script or screenshot for every criterion, never the agent's word.
9. 03.09Testing a browser voice agent: what counts as evidenceVerificationHow AR Labs tests a browser voice agent: what typed sessions and automated tests can prove, what needs a person listening, and how to state the gaps.
10. 03.10Code review gates for AI-generated codeRuleHow AR Labs gates AI-generated code: a pull request per task, a four-axis review by a second model in a fresh context, and a person's merge.
11. 03.11Git worktree isolation for parallel AI coding agentsRuleHow AR Labs isolates parallel AI coding agents with git worktrees: each task on its own branch and checkout, and a hook that blocks edits to main.
12. 03.12Task graphs for AI coding agents with BeadsPracticeHow AR Labs keeps a task graph for AI coding agents in Steve Yegge's Beads, and the rules that make the graph safe to dispatch work from.
13. 03.13Graph engineering for multi-agent systemsFrameworkGraph engineering for AI agent orchestration as AR Labs applies it: nodes, edges, state, routers and gates, and how our coding pipeline follows them.
14. 03.14Harness engineering for AI coding agentsPracticeHarness engineering at AR Labs: each recurring AI coding agent mistake becomes a hook or check, so rules that matter are enforced, not remembered.
15. 03.15Implementation walkthroughsRuleWhy AR Labs' AI coding agents write an implementation walkthrough as they work: a plain account of why each change was made and in what order.
16. 03.16Reuse before createRuleHow AR Labs stops AI coding agents duplicating code: search before writing, re-read before each edit, and sweep for the same bug after a race fix.
17. 03.17Design docs, ADRs and C4 diagrams for AI coding agentsPracticeHow AR Labs documents architecture so people and AI coding agents can read it: a design doc, one-page ADRs, and C4 diagrams kept as text.
18. 03.18Progressive disclosure for AI agent contextPracticeHow AR Labs layers project documents so AI agents and people load the most constraining context first: outcome, strategy, active work, then tasks.
19. 03.19Writing rules for AI agents and humansRuleThe plain-language writing rules AR Labs loads into every AI agent session, why grounding matters most, and how a writing-quality skill audits prose.
20. 03.20Act or ask: when an AI agent should ask permissionRuleHow AR Labs' AI agents decide whether to act and report, or stop and ask a person: by reversibility, ownership and whose decision it is.
21. 03.21Client delivery framework for AI projectsFrameworkThe seven-phase process AR Labs uses to take an AI project from first conversation to a running system, with an evidence gate between phases.
22. 03.22Unattended AI coding agent runsPracticeHow AR Labs runs AI coding agents unattended: one execution contract, facts passed in, typed stops, scripted checks and every deviation logged.
23. 03.23Human-in-the-loop gates for AI coding agentsPracticeHuman-in-the-loop workflow design for AI coding agents at AR Labs: where a person must decide, and how each gate costs one question with a recommendation.
24. 03.24Model routing and agent economicsPracticeHow AR Labs routes each AI agent task to a model and effort level, when a subagent costs more than it saves, and why quality evidence, not price, decides.
25. 03.25Multi-agent orchestration for AI coding agentsPracticeHow AR Labs orchestrates AI coding agents on one body of work: a task graph decides what is ready, each task gets a worktree, overlapping work waits its turn.
26. 03.26Interface contracts between AI agents: who may write whatAgent orchestrationHow AR Labs lets AI agents share state safely: interface contracts drawn between C4 containers, one writer per file, direction down, status up.
27. 03.27Anatomy of a coding agent harnessPracticeThe parts of an AI coding agent harness such as Claude Code (instructions, rules, skills, hooks, subagents, tools, memory) and what each is for at AR Labs.
28. 03.28Understanding code that AI coding agents writePracticeHow AR Labs keeps people able to read code written by AI coding agents: build walkthroughs, guided readings of a change, codebase tours and a code tutor.

The Claude Code skills AR Labs uses for research, planning, review, writing, diagrams and agent orchestration: what each does and where it came from.

1. 04.01Coding Planner: Claude Code skill for spec-driven developmentOrchestrationClaude Code skill AR Labs built to route coding work into its spec-driven pipeline: one task, or a campaign that fans out into tracked sub-tasks.
2. 04.02Grill Me: Claude Code skill for design interviewsPlanningClaude Code skill for design interviews, adapted by AR Labs from Matt Pocock's grill-me: one question per decision, then a design brief agents plan from.
3. 04.03Doc Review: Claude Code skill for reviewing design docsReviewClaude Code skill from AR Labs for reviewing design docs and specs with cross-model reviewers, then a cold reader who tests whether the redraft can be followed.
4. 04.04Write Well: Claude Code skill for AI writing qualityWritingClaude Code skill for AI writing quality, built at AR Labs: diagnoses AI tells, rewrites and scores prose, and checks first that the reader has the context.
5. 04.05ResearchBot: evidence-based research for Claude CodeResearchClaude Code research skill from AR Labs: scores each source with the CRAAP test, triangulates major claims, records conflicts, gaps and confidence.
6. 04.06Multi-agent research: Claude Code skill for web researchResearchClaude Code skill for multi-agent web research, adapted by AR Labs from Daniel Miessler's PAI: 3, 9 or 24 parallel agents, then one sourced report.
7. 04.07Orchestrate: Claude Code orchestration skill for subagentsOrchestrationClaude Code orchestration skill from AR Labs: decide whether a task needs subagents, size the fleet, pick each model, brief each agent, then verify.
8. 04.08To Issues: Claude Code skill to split a plan into tasksPlanningClaude Code skill adapted by AR Labs from Matt Pocock's to-issues: splits a plan into vertical-slice tasks for AI coding agents, wired by dependency.
9. 04.09Taskgraph: Claude Code skill for AI agent status mapsProject statusClaude Code skill from AR Labs for status maps of AI agent work: what is done and where it landed, what runs next, and which decisions wait for a person.
10. 04.10Goal map: Claude Code skill to map AI agents onto goalsVisualisationClaude Code skill from AR Labs that maps AI agent work onto goals: each project on a spiral around the month's goal by priority, its agents as moons.
11. 04.11C4 diagrams: Claude Code skill for architecture diagramsDiagrammingClaude Code skill from AR Labs for C4 architecture diagrams in Mermaid or Structurizr DSL: fixed colours, labelled arrows, a legend and a review checklist.
12. 04.12SVG diagrams: Claude Code skill for architecture diagramsDiagrammingClaude Code skill for SVG architecture and agent-flow diagrams, built at AR Labs: hand-positioned, editable text, light and dark themes, one file.
13. 04.13Paper-and-ink theme: Claude Code skill for design tokensDesign systemClaude Code skill for design tokens, built at AR Labs: one CSS tokens file, a starter page and a written visual language, so agents build pages in one look.
14. 04.14Comprehend: Claude Code skill to understand AI code changesCode comprehensionClaude Code skill from AR Labs for understanding AI-generated code changes: a guided reading of the diff in runtime order, with notes, a diagram and a quiz.
15. 04.15Summary: Claude Code skill to summarise AI agent repliesWritingClaude Code skill from AR Labs for summarising an AI agent's last reply by progressive zoom: the main point in one sentence, then its parts, strictly nested.
16. 04.16Break It Down: Claude Code command to chunk long AI repliesConversation designOpen-source Claude Code slash command from AR Labs that turns an overwhelming AI reply into one question or chunk at a time, sized to human working memory.
17. 04.17Design Probe: Claude Code skill for UI design explorationDesignClaude Code skill from AR Labs for UI design exploration: themed HTML prototypes, desktop and phone screenshots, one change per round, then a locked reference.
18. 04.18Frontend Implementer: Claude Code skill for responsive UIFrontend engineeringClaude Code skill from AR Labs for responsive React and Tailwind UI: finds 13 layout mistakes agents repeat, screenshots five viewports, checks contrast.
19. 04.19Voice sessions: Claude Code skill for talking to AI agentsVoiceClaude Code skill from AR Labs that starts a spoken conversation with an AI voice agent about a document, a design interview or a list of items, in the browser.
20. 04.20Strategy agent skills: planning cadences and escalationStrategyTwo Claude Code skills behind the AR Labs strategy agent: one runs yearly-to-daily planning cadences as a state machine, one files escalations.
21. 04.21Dream: Claude Code skill for agent memory consolidationMemoryClaude Code skill for agent memory consolidation, adapted at AR Labs from OpenClaw's Dreaming: a weekly pass over session transcripts writes a dated record.
22. 04.22Strategic alignment: Claude Code skill for prioritisationPrioritisationClaude Code skill for prioritisation, built at AR Labs: when focus breaks down, list active projects, sort them by value and momentum, and pick one thing.
23. 04.23Advisor mode: Claude Code skill against AI sycophancyConversation modeClaude Code skill against AI sycophancy, built at AR Labs: the agent challenges your decision, rates its confidence and holds firm without new facts.
24. 04.24Building in public: Claude Code skill for technical bloggingContentClaude Code skill AR Labs uses to turn daily engineering work into technical blog articles, with an eight-part framework and the shorter posts pre-written.
25. 04.25Create Skill: Claude Code skill for writing agent skillsSkill authoringClaude Code skill for writing agent skills, adapted at AR Labs from Daniel Miessler's PAI: folder structure, trigger descriptions and traceable imports.
26. 04.26Beads skills: Claude Code skills for the Beads issue trackerWork trackingTwo Claude Code skills AR Labs built for the Beads issue tracker: one routes tracker operations through our policy, one links a session to its task.
27. 04.27Walkthrough: Claude Code skill for codebase walkthroughsCode comprehensionClaude Code skill for codebase walkthroughs, packaged by AR Labs from Simon Willison's Showboat method: every code excerpt is extracted from the real files.
28. 04.28Code Tutor: Claude Code commands for learning a codebaseTeachingClaude Code commands from AR Labs for learning code by building it: an agent writes a tutorial on the feature, then tutors you Socratically as you code.

Dated, source-evaluated research reports from AR Labs on voice AI, AI agent orchestration, context engineering, RAG, verification and web platforms.

1. 05.01AI research agent methodology: how AR Labs evaluates sourcesMethodHow AR Labs runs research with AI agents: depth levels, CRAAP source scoring, triangulation, recorded conflicts and gaps, and confidence grades.
2. 05.02Wispr Flow alternatives: local speech-to-text apps for macOSVoice AIWhich macOS dictation app gives low cost, on-device privacy and system-wide typing? Free and one-time-purchase Whisper-based apps match Wispr Flow.
3. 05.03Pipecat voice agent architecture: pipeline, STT, TTS and latencyVoice AIHow to build a real-time voice agent on Pipecat with Daily WebRTC, Deepgram, ElevenLabs and LangGraph: configuration, latency budgets and known pitfalls.
4. 05.04Sesame Conversational Speech Model (CSM) and open-source voice AIVoice AIHow Sesame's CSM produces natural conversational speech, and which open-source frameworks, speech-to-speech and TTS models can build comparable voice agents.
5. 05.05Real-time voice AI agent architecture for coding agents such as Claude CodeVoice AIHow to put a real-time voice interface in front of a CLI coding agent like Claude Code: multimodal models vs STT-LLM-TTS pipelines, transports, VAD and bridges.
6. 05.06Building autonomous agents on Hermes: planning, cron and resumeAgent orchestrationHow public builders make Hermes agents plan, run over many sessions and resume. Long runs rely on cron and Kanban, not a plan the agent revises across weeks.
7. 05.07AI agent planning and execution loop: holding a plan, choosing the next task and resumingAgent orchestrationHow an AI agent should hold a plan, pick its next task among competing goals and resume after a break, drawn from marketing planning and harness design.
8. 05.08Logical levels and the theory of types applied to AI agent goalsAgent orchestrationWhat Bateson's logical levels and Russell's theory of types imply for nesting an AI agent's goals, approaches and tasks, and which failures to avoid.
9. 05.09AI software factories: a survey of backlog-to-merged-code systemsAgent orchestrationA survey of more than 60 AI software factories that turn a backlog or spec into verified, merged code. Deploy stages and merge queues are rarest.
10. 05.10AI agent orchestration driver: reconciliation loops versus eventsAgent orchestrationWhich execution model suits a long-running, crash-prone AI agent orchestration driver? A level-triggered reconciler that re-reads durable state.
11. 05.11Prompting multi-agent orchestration: delegation rules and failure modesAgent orchestrationHow should a lead AI agent brief and coordinate parallel worker agents? Anthropic's published practice plus nine recurring failure modes and their fixes.
12. 05.12Harness engineering for AI agents: origins and sourcesAgent orchestrationWhere did the term harness engineering come from? Hashimoto coined the practice and LangChain the formula Agent = Model + Harness, in early 2026.
13. 05.13Self-hosted always-on AI agents: OpenClaw, Hermes, Claude Agent SDKAgent orchestrationSelf-hosted always-on AI agent platforms in April 2026: OpenClaw, Hermes, NanoBot, NanoClaw, OpenFang and the Claude Agent SDK compared on security and memory.
14. 05.14Moving AI agent sessions between Claude Code, Codex, Pi and machinesContext and memoryHow to carry an AI agent session across harnesses and machines. Open-source converters cover much of it, but none moves every native record losslessly.
15. 05.15AI agent memory for changing facts: append-only records versus memory systemsContext and memoryWhat memory lets an AI agent know which facts hold now and how they changed. Evidence favours curated, dated records read in full over extracted memory.
16. 05.16Markdown RAG for Claude Code and LangGraph: grep versus vector searchContext and memoryHow to run RAG over a few markdown files from Claude Code and port it to LangGraph: grep retrieval first, a ChromaDB vector script only for semantic queries.
17. 05.17RAG frameworks for LangGraph and Claude Code in 2026Context and memoryWhich RAG frameworks, vector stores, PDF parsers and embeddings suit document Q&A with LangGraph? LangChain plus LangGraph, hybrid search and Qdrant lead.
18. 05.18Context engineering for long-running AI agents: strategies and memoryContext and memoryHow to keep long-running AI agents effective as context fills: write, select, compress and isolate context, plus memory, checkpointing and AI review.
19. 05.19Test budgets for AI coding agents: cost proportional to change sizeVerificationHow teams scale verification to change size: test impact analysis, staged CI and a ten-minute gate. No AI coding agent found ties tests to change size.
20. 05.20Why multi-round LLM code review loops divergeVerificationWhy repeated LLM code review rounds find more critical issues on smaller problems instead of converging, and which mitigations have the strongest evidence.
21. 05.21LLM-as-reviewer methods: quality and readability checksVerificationWhich LLM-as-reviewer methods lift quality and readability: fresh-context review, per-criterion rubrics, cross-family judges. Readability scores mislead.
22. 05.22Displaying a task dependency graph so one person sees what to do nextInterfacesHow to draw a 20 to 60 item work dependency DAG so one person sees what to do next and why. Layered swimlanes with a ready-to-start band came out ahead.
23. 05.23Scheduling a task dependency graph across AI coding agentsAgent orchestrationHow to run a task dependency graph on a small pool of AI coding agents: scheduling theory favours ready-set list scheduling by critical path over waves.
24. 05.24Technical blog readability: evidence-based typography and layoutInterfacesWhich typography, contrast, spacing and code-block choices make technical blog posts easiest to read in light and dark mode, graded by strength of evidence.
25. 05.25Conversational navigation models for LLM voice agentsInterfacesHow dialogue systems escape rigid linear question flows: agenda and stack navigation, information-state extraction, and LLM tools such as pipecat-flows.
26. 05.26Node-graph and infinite-canvas tools for a human and AI agent task graphInterfacesWhich node-graph libraries and infinite-canvas tools suit a shared human and AI agent view of a task dependency graph? React Flow and beads_viewer fit best.
27. 05.27Cloudflare Workers, D1 and R2 as a default for new web appsPlatformsWhy developers default to Cloudflare Workers, D1 and R2 for new web apps, and which critiques hold: single-region writes, runtime limits, lock-in, no spend cap.
28. 05.28Technical blog stacks: how Karpathy, Simon Willison, Dan Luu and Julia Evans build theirsPlatformsHow seven well-known technical writers build their blogs. Simple, stable infrastructure and plain Markdown go with more posting and easier migrations.
29. 05.29Vibe coding to vibe engineering: AI-assisted coding in early 2026MethodsHow AI-assisted coding moved from casual vibe coding to structured vibe engineering, and the tools, methods and multi-agent patterns of early 2026.
30. 05.30Writing PRDs and PRPs for AI coding agents: expert methods comparedMethodsHow experts write PRDs and PRPs (product requirement prompts) for AI coding agents: shorter documents, dependency-ordered phases and testable checkpoints.

AR Labs case studies: an anonymised client design engagement, written up as engineering work from discovery to design.

1. 06.01Music school enquiry and scheduling automationProduct designAn anonymised AR Labs case study: enquiry capture, teacher matching, trial booking and re-placement for a music school, with the owner approving each step.
2. 06.02Business process discovery for workflow automationMethodAR Labs' business process discovery method: screen recordings, AI frame analysis and layered evidence show how a small business works before automation.
3. 06.03Scheduling workspace design for teacher matchingInterface designAR Labs' interface design for music school teacher matching: a teacher's week beside ranked students, and a rule for which data it owns or mirrors.

How AR Labs runs as a small company of one person and a set of AI agents: planning cadences, a strategy agent, a file-based agent inbox and dashboards.

1. 07.01Strategy agent planning cadencesOperating methodHow AR Labs runs a strategy agent: yearly, quarterly, monthly, weekly and daily planning cadences that link long-range direction to daily focus and escalation.
2. 07.02Running a one-person company with AI agents: departmentsOperating methodHow AR Labs organises one person's work into five departments and named activity streams, so a strategy agent can allocate time and agents can take roles.
3. 07.03AI agent operations dashboards and goal mapsDashboardAI agent operations dashboard design from AR Labs: what a person must see to supervise agents, and the goal map, week page and cadence dashboard we plan with.
4. 07.04Monitoring multiple AI agent sessions from one screenOperating methodWhat AR Labs thinks a person needs to see and steer many AI agent sessions at once, and the rules we use for returning to a conversation safely.
5. 07.05Escalation workflow for AI agentsOperating methodThe escalation workflow AR Labs uses for AI agents: typed, tracked escalations with structured notes, a decision lifecycle and a five-level authority model.
6. 07.06File-based inbox protocol for AI agentsOperating methodThe file-based inbox protocol AR Labs uses so independent AI agents exchange requests, replies and attachments as durable files, with chat used only to nudge.
7. 07.07A git repository as a second brain for AI coding agentsOperating methodHow AR Labs uses one git repository as memory for AI coding agents: layered documents, one per concern, and one instruction file for Claude Code and Codex.
8. 07.08Session history and memory for AI coding agentsOperating methodHow AR Labs gives stateless Claude Code sessions a memory: event logs written by hooks, history search, sessions linked to tasks, weekly consolidation.
9. 07.09AI scheduling for Google Calendar by domains and modesOperating methodThe conversational AI scheduler AR Labs uses to turn weekly priorities into Google Calendar blocks by domains, modes and containers, and to recover lost time.
10. 07.10How AR Labs demos use data, and their limitsData useWhat the AR Labs Salesbot voice demo collects, where it is stored, which outside services process it, and what the demo is not. A statement of fact.

How-to guides from AR Labs for small service business owners who want to fix or automate a workflow, starting with the simplest change before software or AI.

1. 08.01How to automate music school scheduling from enquiry to booked lessonHow-to guideAn AR Labs guide for music school owners: capture what a teacher match needs, shortlist teachers, book trial lessons and automate the clerical work.
2. 08.02How to automate tutor matching and trial lesson bookingHow-to guideAn AR Labs guide for tutoring companies: turn enquiries and assessments into tutor matches and booked first sessions, and know when your platform is enough.
3. 08.03What to ask on a music lesson or tutoring enquiry formHow-to guideAn AR Labs reference for music schools and tutoring businesses: what to ask on an enquiry form, and why each field matters for matching and trial booking.
4. 08.04How to turn service call enquiries into dispatch-ready jobsHow-to guideAn AR Labs guide for plumbing, HVAC and field-service firms: call scripts, forms and software settings so every job is dispatch-ready without call-backs.
5. 08.05How to automate client intake for an agency or consultancyHow-to guideAn AR Labs guide for agencies and consultancies: turn a vague project enquiry into a confirmed brief before the first call, with forms you may already have.
6. 08.06How to automate new client onboarding for a bookkeeping practiceHow-to guideAn AR Labs guide for bookkeeping practices: collect a new client's details, documents and software access with one request and reminders, never passwords.
7. 08.07How to follow up enquiries and quotes without losing themHow-to guideAn AR Labs guide for small service businesses that lose enquiries and quotes: one list, a next action and owner on each, a routine, then CRM reminders.
8. 08.08How to map a business process before automating itHow-to guideAn AR Labs guide for small business owners: map a workflow as it really runs, count what fails, keep the human decisions, then pick the smallest fix.
9. 08.09How to collect client facts before a first legal consultationHow-to guideAn AR Labs guide for small law firms: collect a prospective client's facts before the first consultation, conflict check first, no advice, a person deciding.
10. 08.10How to decide what to automate with AI in a small businessHow-to guideAn AR Labs guide to deciding what to automate with AI in a small business: measure costly workflows, pick process, software, rules or AI, test one change.