# The Third Way Company > The third way between a full dev team and a freelancer marketplace: senior engineers directing specialized AI agents to ship production software, automate workflows, and upskill teams — automation for the repetitive, humans for the judgment. Founded by Fabs. Expanded index with descriptions: https://thethirdwaycompany.com/llms-full.txt ## Pages - [Home](https://thethirdwaycompany.com): what we do and how we work - [Our thesis](https://thethirdwaycompany.com/thesis): why the third way beats both staffing models - [About](https://thethirdwaycompany.com/about): who we are - [Services index](https://thethirdwaycompany.com/services) and [case studies index](https://thethirdwaycompany.com/work) - [Compare](https://thethirdwaycompany.com/compare): us vs hiring vs freelancer marketplaces - [Pricing](https://thethirdwaycompany.com/pricing): engagement models with transparent price bands for assessments, builds, and upskilling — fixed per milestone or retainer, never an open meter - [Contact](https://thethirdwaycompany.com/contact) — or email hello@thethirdwaycompany.com - [Security](https://thethirdwaycompany.com/security) · [Privacy](https://thethirdwaycompany.com/privacy) · [Terms](https://thethirdwaycompany.com/terms) · [Cookies](https://thethirdwaycompany.com/cookies) · [Subprocessors](https://thethirdwaycompany.com/subprocessors) ## Services - [Software Development](https://thethirdwaycompany.com/services/software-development): Production-ready applications built by senior engineers directing AI agents. Idea to launch in weeks, with code your team can own. - [AI Workflow Automation](https://thethirdwaycompany.com/services/ai-workflows): We audit your team's repetitive workflows and automate them with AI agents — measurable efficiency gains and lower operational spend. - [AI Upskilling Program](https://thethirdwaycompany.com/services/ai-upskilling): Hands-on training that turns your developers and operators into confident AI users. Practical skills, real workflows, measurable results. - [AI Risk and Security Assessment](https://thethirdwaycompany.com/services/ai-risk-security-assessment): Find the AI risks in your organization before auditors and attackers do — shadow AI, data leakage, and compliance gaps, with a concrete remediation plan. - [Agent Ship Week](https://thethirdwaycompany.com/services/agent-ship-week): A fixed £5,000 five-day sprint: opportunities mapped before day one, constraints encoded before behavior, one AI agent live in production by day five. ## Industries - [Industries index](https://thethirdwaycompany.com/industries): the three services applied to one vertical at a time, each with a proof point from real engagements - [Law firms](https://thethirdwaycompany.com/industries/law-firms): How law firms adopt AI without breaching confidentiality: a Claude plugin for your practice’s admin and drafts, workflow automation for the routine, and upskilling — nothing leaves the firm without a lawyer’s OK. - [Financial services](https://thethirdwaycompany.com/industries/financial-services): Fintech and financial-services teams ship customer-facing products in weeks, automate back-office workflows, and upskill engineers — senior engineers directing AI agents, a human author on every regulated edge. - [Logistics](https://thethirdwaycompany.com/industries/logistics): Logistics operators automate exception handling, document chasing, and manual reporting with AI agents — one mid-market operator got 900+ hours a month back and cut processing errors 80%. ## Case studies - [From concept to production MVP in six weeks](https://thethirdwaycompany.com/work/fintech-mvp-six-weeks): A funded fintech needed a customer-facing product live before their next board meeting. Traditional agency quotes: 5–6 months. We shipped in six weeks. - [Reclaiming 900 hours a month from manual operations](https://thethirdwaycompany.com/work/logistics-operations-automation): A twelve-person operations team was drowning in manual exception handling and copy-paste reporting. We automated the top three workflows and gave them 900 hours a month back. - [Finding 60 unsanctioned AI tools before the auditors did](https://thethirdwaycompany.com/work/professional-services-shadow-ai): A 400-person firm had an AI policy, a procurement process, and — as we found — sixty AI tools in use that neither knew about. Including three sending client data offshore. ## Writing - [Writing index](https://thethirdwaycompany.com/blog): AI governance, certification guides, and delivery practice — newest first - [A free Beginner course on Claude Code — the habits, not the commands](https://thethirdwaycompany.com/blog/claude-code-beginner-course) (2026-10-10): Every Claude Code tutorial teaches the commands, and the commands change every month. Our free Beginner course — five lessons, a capstone, three electives — teaches the habits that survive the updates. - [Everyone uses AI. Almost nobody gets returns.](https://thethirdwaycompany.com/blog/the-5-percent-problem) (2026-10-07): Two 2026 research reports arrive at the same paradox from opposite directions. Adoption is nearly universal, returns are vanishingly rare — and the minority that gets them changed the work, not the shopping list. - [What the third way actually is](https://thethirdwaycompany.com/blog/what-the-third-way-actually-is) (2026-10-05): “AI enablement” has become a bucket everyone sells into. We are not bidding for that word. The third way is a delivery model with four tests — and every one of them is a mechanic our delivery board enforces, not a value we aspire to. - [Agents don’t need memory, they need documentation](https://thethirdwaycompany.com/blog/documentation-over-memory) (2026-10-05): A developer who spent a year building agent-memory tooling concluded the category is flawed: agents don’t need recall of past chats, they need documents they can read and you can audit. Here is what that means for a firm that writes no code. - [What your always-on agent wakes you for](https://thethirdwaycompany.com/blog/always-on-agents) (2026-09-30): The launch post says the quiet part out loud: the agent “comes back when it needs a decision.” That sentence is the whole design problem. - [Nine famous websites, zero agent-ready](https://thethirdwaycompany.com/blog/agent-ready-check) (2026-09-29): Stripe scored 35. The New York Times 44. Anthropic — the company that makes Claude — 30. Here’s what the numbers actually mean, and how to read your own. - [Your 10-page system prompt is a smell](https://thethirdwaycompany.com/blog/context-over-prompts) (2026-09-29): The best models now reason from context. Most instruction bloat is information that belongs in retrieval, memory, or tools — not prose. - [Your next customer might be an AI agent](https://thethirdwaycompany.com/blog/visible-to-ai-agents) (2026-09-24): An insurance executive says a shopping agent strips the legal details out of quotes. Most brands don’t even know they’re being visited. - [Falling token prices won’t fix your AI budget](https://thethirdwaycompany.com/blog/falling-token-prices) (2026-09-23): The unit price of machine-generated work keeps falling. The expensive parts of using AI were never the generation. - [Claude Developer Foundations (CCDV-F) exam prep](https://thethirdwaycompany.com/blog/claude-developer-foundations-exam-prep) (2026-09-20): Of Anthropic's three Claude certifications, the Developer one is the fair one — if you already build with the API, you are most of the way there. Here is the one-to-two-week plan I wrote for a friend who asked for the fastest path. - [Claude Architect Professional (CCAR-P) exam prep](https://thethirdwaycompany.com/blog/claude-architect-professional-exam-prep) (2026-09-20): I passed Anthropic's Architect Professional exam with zero preparation, booked three hours beforehand. Not a flex — a statement about what the exam tests: org-level design judgment, governance, and knowing where AI fits. - [Claude Architect Foundations (CCAR-F) exam prep](https://thethirdwaycompany.com/blog/claude-architect-foundations-exam-prep) (2026-09-20): Anthropic's "Foundations" architect exam is the hardest of the three Claude certifications — more technical than the Developer one. I passed it; here is the prep plan I would follow, down to the only mock exams at real exam difficulty. - [Your law firm could have its own Claude plugin](https://thethirdwaycompany.com/blog/claude-plugins-for-law-firms) (2026-09-20): Claude now runs books and chases invoices for small businesses — but not for law firms. We build the missing version: one that knows your matters, drafts your routine work, and sends nothing without a lawyer’s OK. - [Jev and typesafe AI: judgment, not text](https://thethirdwaycompany.com/blog/jev-and-typesafe-ai) (2026-09-19): Most "AI feature" plans end at the same place: prompt a chatbot, ask for JSON, and write a parser robust enough to survive the days it doesn’t. Jev — TypeSafe AI’s first System One model — does something different: it returns typed judgments and probabilities your code can branch on. - [Shadow AI: the security risk already inside your company](https://thethirdwaycompany.com/blog/shadow-ai-security-risk) (2026-09-11): In every assessment we run, we find AI tools the security team did not know existed. Shadow AI is not a future risk — it is a present, unaudited data exfiltration channel. - [What we actually let AI agents do (and what we never will)](https://thethirdwaycompany.com/blog/what-we-let-ai-agents-do) (2026-09-04): "AI-assisted development" is uselessly vague. Here is our actual autonomy policy: the exact lines we draw, and why they sit where they do. - [Human-only vs. AI-only vs. the third way: a cost model](https://thethirdwaycompany.com/blog/human-only-vs-ai-only-cost-model) (2026-08-26): Everyone has an opinion on AI and software teams. Fewer people have run the numbers. Here is the cost model we use with clients, derived from real engagements. - Blog feed (Atom): https://thethirdwaycompany.com/feed.xml — every post above, machine-readable ## Glossary - [Agentic-AI glossary](https://thethirdwaycompany.com/glossary): hand-curated, plain-English definitions of the agentic-AI vocabulary — MCP, llms.txt, shadow AI, agent readiness, A2A, agent cards, context engineering, model routing, human-in-the-loop, and more — each with why it matters for buyers - [Model Context Protocol (MCP)](https://thethirdwaycompany.com/glossary/model-context-protocol): An open protocol that connects AI applications to external tools and data through one standard interface — so integrations are written once and used by any capable client. - [llms.txt](https://thethirdwaycompany.com/glossary/llms-txt): A markdown file at a site's root that indexes its content for AI models — robots.txt's cousin, written for machines that read rather than follow links. - [Shadow AI](https://thethirdwaycompany.com/glossary/shadow-ai): Employees using AI tools their organization never approved, with data it would never deliberately send anywhere — the most common finding in AI security assessments. - [Agent readiness](https://thethirdwaycompany.com/glossary/agent-readiness): How well a public website serves AI agents as visitors — structured facts, machine-readable signals, and the emerging protocols agents can act through. - [Agent2Agent Protocol (A2A)](https://thethirdwaycompany.com/glossary/agent2agent-protocol): An open protocol for AI agents to discover each other, exchange tasks over JSON-RPC, and return results — across frameworks and vendors. - [Agent card](https://thethirdwaycompany.com/glossary/agent-card): A machine-readable JSON file at a well-known URL describing what an agent can do and how to reach it — a business card other agents can parse. - [Context engineering](https://thethirdwaycompany.com/glossary/context-engineering): Designing what an AI model actually sees — retrieval, memory, tools, user state — instead of writing ever-longer instructions. The discipline that succeeded prompt engineering. - [Model routing](https://thethirdwaycompany.com/glossary/model-routing): Directing each piece of work to the AI model that fits it — by capability, cost, and speed — with explicit rules for when work escalates. - [Human-in-the-loop](https://thethirdwaycompany.com/glossary/human-in-the-loop): Deliberately keeping a person at the decision points of automated work — approvals, escalations, and final accountability — while agents do the labor around them. - [Agentic commerce](https://thethirdwaycompany.com/glossary/agentic-commerce): Commerce where AI agents search, compare, and buy on people's behalf — and where a business has to be legible to a buyer that never sees the page. - [Agent skills](https://thethirdwaycompany.com/glossary/agent-skills): Markdown instruction packages served at a well-known URL that teach visiting AI agents how to use a site — documentation agents load exactly when they need it. - [Web Bot Auth](https://thethirdwaycompany.com/glossary/web-bot-auth): An emerging web standard that lets AI agents identify themselves cryptographically — HTTP requests signed with published keys, so "which agent is this?" stops being guesswork. - [Evals](https://thethirdwaycompany.com/glossary/evals): Systematic tests for AI systems — fixed inputs, graded outputs — that measure whether a change made things better or worse. The safety net under everything else in this glossary. - [Tool use (function calling)](https://thethirdwaycompany.com/glossary/tool-use): The mechanism by which an AI model calls external functions — declared with typed schemas — instead of only producing text. The difference between an assistant and an agent. - [System prompt](https://thethirdwaycompany.com/glossary/system-prompt): The standing instruction layer that shapes every response a model gives — role, boundaries, format. Where preferences belong; where information does not. ## Study guides - [Study guides for the Claude certifications](https://thethirdwaycompany.com/guides): the index — free, unofficial guides for CCDV-F, CCAR-F and CCAR-P (CCDV-F and CCAR-F also in French), the Claude Code course, plus field guides on the questions clients ask next - [Can an AI agent buy from you?](https://thethirdwaycompany.com/guides/agentic-commerce): field guide to agentic commerce — what Instant Checkout, ACP, UCP and AP2 actually are, the Amazon v. Perplexity ruling, who needs agent-payable surfaces today (retail, travel, B2B self-serve), and an eight-item seller checklist - [Claude Developer Foundations (CCDV-F) study guide](https://thethirdwaycompany.com/guides/ccdv-f): eight domain-by-domain guides to Anthropic's developer certification — exam facts, module coverage, and the traps - [Start here — CCDV-F primer](https://thethirdwaycompany.com/guides/ccdv-f/start-here): New to Claude? The primer that readies you for the eight domain guides: what Claude is, what an AI engineer does, and your first API call on paper. - [Agents and workflows — CCDV-F domain 01](https://thethirdwaycompany.com/guides/ccdv-f/agents-and-workflows): Who decides the next step — your code or the model? The 14.7% of the CCDV-F exam on agentic patterns, the agent loop, tool-use loops and subagents. - [Applications and integration — CCDV-F domain 02](https://thethirdwaycompany.com/guides/ccdv-f/applications-and-integration): A third of the CCDV-F exam: the Messages API in depth, model lifecycles, the REST surface, endpoints, headers, batches and configuration scopes. - [Claude Code — CCDV-F domain 03](https://thethirdwaycompany.com/guides/ccdv-f/claude-code): CLAUDE.md hierarchy, rules, auto memory, settings.json precedence, AGENTS.md fallback rules, skills, subagents and headless mode — the complete Claude Code slice of CCDV-F. - [Eval, testing, and debugging — CCDV-F domain 04](https://thethirdwaycompany.com/guides/ccdv-f/evals-testing-and-debugging): Error types, retries and the three things that look like a rate limit — the smallest CCDV-F domain, and the debugging kit you use everywhere else. - [Model selection and optimization — CCDV-F domain 05](https://thethirdwaycompany.com/guides/ccdv-f/model-selection-and-optimisation): Fable, Opus, Sonnet or Haiku — effort levels, thinking modes, token pricing, counting and cost modelling for the 16.8% model-selection slice of CCDV-F. - [Prompt and context engineering — CCDV-F domain 06](https://thethirdwaycompany.com/guides/ccdv-f/prompt-and-context-engineering): Context rot, compaction, few-shot examples, XML structure, constrained decoding and strict tool use — the prompt and context engineering domain of CCDV-F. - [Security and safety — CCDV-F domain 07](https://thethirdwaycompany.com/guides/ccdv-f/security-and-safety): Prompt injection, layered guardrails, hooks, key hygiene and workload identity — the CCDV-F security domain, where the wrong answers sound sensible. - [Tools and MCPs — CCDV-F domain 08](https://thethirdwaycompany.com/guides/ccdv-f/tools-and-mcp): Tool schemas and descriptions, MCP primitives and transports, and choosing between built-ins, custom tools, MCP servers and skills for CCDV-F. - [Guide d’étude CCDV-F en français](https://thethirdwaycompany.com/guides/ccdv-f/fr): Un guide d’étude gratuit, domaine par domaine, pour l’examen CCDV-F — ce que chacun des huit domaines teste, les faits qui reviennent et les pièges à éviter. - [Commencez ici — CCDV-F primer (français)](https://thethirdwaycompany.com/guides/ccdv-f/fr/start-here): Vous débutez avec Claude ? La mise en route qui vous prépare aux huit guides de domaine : ce qu’est Claude, ce que fait un ingénieur IA au quotidien et votre premier appel d’API sur le papier. - [Agents et workflows — CCDV-F domain 01 (français)](https://thethirdwaycompany.com/guides/ccdv-f/fr/agents-and-workflows): Qui décide de la prochaine étape — votre code ou le modèle ? Les 14,7 % de l’examen CCDV-F consacrés aux schémas agentiques, à la boucle de l’agent, aux boucles d’utilisation des outils et aux sous-agents. - [Applications et intégration — CCDV-F domain 02 (français)](https://thethirdwaycompany.com/guides/ccdv-f/fr/applications-and-integration): Un tiers de l’examen CCDV-F : l’API Messages en profondeur, cycles de vie des modèles, surface REST, endpoints, en-têtes, batches et portées de configuration. - [Claude Code — CCDV-F domain 03 (français)](https://thethirdwaycompany.com/guides/ccdv-f/fr/claude-code): Hiérarchie de CLAUDE.md, fichiers de règles, mémoire automatique, précédence de settings.json, règles de repli d’AGENTS.md, skills, sous-agents et mode headless — toute la tranche Claude Code du CCDV-F. - [Évals, tests et débogage — CCDV-F domain 04 (français)](https://thethirdwaycompany.com/guides/ccdv-f/fr/evals-testing-and-debugging): Types d’erreurs, retries et les trois choses qui ressemblent à une limite de débit — le plus petit domaine du CCDV-F, et la trousse de débogage que vous utiliserez partout ailleurs. - [Sélection et optimisation de modèles — CCDV-F domain 05 (français)](https://thethirdwaycompany.com/guides/ccdv-f/fr/model-selection-and-optimisation): Fable, Opus, Sonnet ou Haiku — niveaux d’effort, modes de réflexion, tarification des tokens, comptage et modélisation des coûts pour les 16,8 % du CCDV-F consacrés à la sélection de modèle. - [Ingénierie du prompt et du contexte — CCDV-F domain 06 (français)](https://thethirdwaycompany.com/guides/ccdv-f/fr/prompt-and-context-engineering): Context rot, compaction, exemples few-shot, structure XML, décodage contraint et tool use strict — le domaine d’ingénierie du prompt et du contexte du CCDV-F. - [Sécurité et sûreté — CCDV-F domain 07 (français)](https://thethirdwaycompany.com/guides/ccdv-f/fr/security-and-safety): Injection de prompt, garde-fous en couches, hooks, hygiène des clés et identité de charge — le domaine sécurité du CCDV-F, là où les mauvaises réponses paraissent sensées. - [Outils et MCP — CCDV-F domain 08 (français)](https://thethirdwaycompany.com/guides/ccdv-f/fr/tools-and-mcp): Schémas et descriptions d’outils, primitives et transports MCP et le choix entre intégrés, outils personnalisés, serveurs MCP et skills pour le CCDV-F. - [Claude Architect Foundations (CCAR-F) study guide](https://thethirdwaycompany.com/guides/ccar-f): five domain-by-domain guides to Anthropic's architect certification — the agent loop, tool design, Claude Code configuration, structured output, context management, and the traps - [Start here — CCAR-F primer](https://thethirdwaycompany.com/guides/ccar-f/start-here): New to the architect track? The primer that readies you for the five domain guides: what CCAR-F actually tests, how it differs from the developer exam, and how to study for it. - [Agentic architecture and orchestration — CCAR-F domain 01](https://thethirdwaycompany.com/guides/ccar-f/agentic-architecture): CCAR-F Domain 1 (27%): the agent loop, when an agent beats a fixed workflow, coordinator–subagent systems, context passing, enforcement and handoff patterns, SDK hooks, task decomposition, session state. - [Tool design and MCP integration — CCAR-F domain 02](https://thethirdwaycompany.com/guides/ccar-f/tool-design-mcp): CCAR-F Domain 2 (18%): what makes a good tool definition, structured error responses, distributing tools across agents and tool choice, MCP servers, transports and scopes, and the built-in tools. - [Claude Code configuration and workflows — CCAR-F domain 03](https://thethirdwaycompany.com/guides/ccar-f/claude-code-configuration): CCAR-F Domain 3 (20%): the CLAUDE.md hierarchy and imports, settings precedence, custom slash commands and skills, path-specific rules, plan mode, iterative refinement and headless CI/CD. - [Prompt engineering and structured output — CCAR-F domain 04](https://thethirdwaycompany.com/guides/ccar-f/prompt-engineering-structured-output): CCAR-F Domain 4 (20%): prompts with explicit criteria, few-shot examples, schema-enforced structured output, validation and retry loops, batch processing, and multi-instance multi-pass review. - [Context management and reliability — CCAR-F domain 05](https://thethirdwaycompany.com/guides/ccar-f/context-management-reliability): CCAR-F Domain 5 (15%): keeping long sessions coherent — compaction, context editing, the memory tool — plus escalation patterns, error propagation in multi-agent systems, codebase exploration, human review and provenance. - [Evidence and transparency — CCAR-F primer](https://thethirdwaycompany.com/guides/ccar-f/evidence): Every claim in this CCAR-F guide, sorted by evidence: the official domain weights and six scenario contexts confirmed against Anthropic’s exam guide, what a real sitting cross-validated, and what is our own inference. - [Guide d’étude CCAR-F en français](https://thethirdwaycompany.com/guides/ccar-f/fr): Un guide d’étude gratuit, domaine par domaine, pour l’examen CCAR-F — ce que chacun des cinq domaines teste, les faits qui reviennent et les pièges à éviter. - [Commencez ici — CCAR-F primer (français)](https://thethirdwaycompany.com/guides/ccar-f/fr/start-here): Vous arrivez sur la filière architecte ? La mise en route qui vous prépare aux cinq guides de domaine : ce que le CCAR-F teste vraiment, comment il diffère de l’examen développeur, et dans quel ordre l’étudier. - [Architecture agentique et orchestration — CCAR-F domain 01 (français)](https://thethirdwaycompany.com/guides/ccar-f/fr/agentic-architecture): Domaine 1 du CCAR-F (27 %) : la boucle de l’agent, quand un agent bat un flux fixe, les systèmes coordinateur–subagents, le passage de contexte, l’imposition et les passations, les hooks du SDK, la décomposition des tâches, l’état de session. - [Conception d’outils et intégration MCP — CCAR-F domain 02 (français)](https://thethirdwaycompany.com/guides/ccar-f/fr/tool-design-mcp): Domaine 2 du CCAR-F (18 %) : ce qui fait une bonne définition d’outil, les réponses d’erreur structurées, la distribution des outils entre agents et le tool choice, les serveurs MCP, transports et portées, et les outils intégrés. - [Configuration et flux de travail Claude Code — CCAR-F domain 03 (français)](https://thethirdwaycompany.com/guides/ccar-f/fr/claude-code-configuration): Domaine 3 du CCAR-F (20 %) : la hiérarchie CLAUDE.md et ses imports, la précédence des settings, commandes slash et skills personnalisés, règles par chemin, mode plan, raffinement itératif et CI/CD headless. - [Ingénierie de prompts et sortie structurée — CCAR-F domain 04 (français)](https://thethirdwaycompany.com/guides/ccar-f/fr/prompt-engineering-structured-output): Domaine 4 du CCAR-F (20 %) : prompts à critères explicites, few-shot, sortie imposée par schéma, boucles de validation et de reprise, traitement par lots, et revue multi-instance multi-passe. - [Gestion du contexte et fiabilité — CCAR-F domain 05 (français)](https://thethirdwaycompany.com/guides/ccar-f/fr/context-management-reliability): Domaine 5 du CCAR-F (15 %) : garder des sessions longues cohérentes — compaction, context editing, memory tool — plus les échelles d’escalade, la propagation des erreurs entre agents, l’exploration de grosses bases de code, la calibration de confiance et la traçabilité. - [Preuves et transparence — CCAR-F primer (français)](https://thethirdwaycompany.com/guides/ccar-f/fr/evidence): Chaque affirmation de ce guide CCAR-F, triée par preuve : les pondérations officielles des domaines et les six contextes de scénario confirmés contre le guide d’examen d’Anthropic, ce qu’une vraie session a recoupé, et ce qui relève de notre propre inférence. - [Claude Architect Professional (CCAR-P) study guide](https://thethirdwaycompany.com/guides/ccar-p): seven domain-by-domain guides to Anthropic's professional architect certification — solution design, model selection, integration and RAG, evaluation, governance, stakeholders, and the traps - [Start here — CCAR-P primer](https://thethirdwaycompany.com/guides/ccar-p/start-here): New to the Architect Professional exam? The primer that readies you for the seven domain guides: what CCAR-P actually tests, how it differs from Foundations, and how to study for a judgment exam. - [Solution design and architecture — CCAR-P domain 01](https://thethirdwaycompany.com/guides/ccar-p/solution-design): CCAR-P Domain 1 (17%): turning business problems into Claude-based solutions, end-to-end architectures with feedback loops, workflow vs agentic patterns, multi-agent orchestration, decomposition, and business value pillars. - [Claude models, prompting and context engineering — CCAR-P domain 02](https://thethirdwaycompany.com/guides/ccar-p/models-prompting-context): CCAR-P Domain 2 (13%): selecting Claude models by trade-off, system prompts and guardrails, zero-shot / few-shot / chain-of-thought, context window budgets and token management, and reuse via prompt caching, modular prompts and Skills. - [Integration — CCAR-P domain 03](https://thethirdwaycompany.com/guides/ccar-p/integration): CCAR-P Domain 3 (19%), the heaviest: capability bloat and least privilege, authentication and authorization gaps, accuracy–latency trade-offs, observability at scale, RAG pipeline design, retrieval strategy by data shape, MCP vs API/CLI vs agent-to-agent, progressive discovery. - [Evaluation, testing and optimization — CCAR-P domain 04](https://thethirdwaycompany.com/guides/ccar-p/evaluation-testing-optimization): CCAR-P Domain 4 (16%): defining metrics across accuracy, latency, cost, safety and security; evaluation datasets and mixed methodologies; A/B testing; diagnosing prompt failures, hallucinations and model mismatch; token and cost optimization; production monitoring. - [Governance, safety and risk management — CCAR-P domain 05](https://thethirdwaycompany.com/guides/ccar-p/governance-safety-risk): CCAR-P Domain 5 (14%): layered guardrails and safety controls, LLM failure modes from hallucination to prompt injection, human-in-the-loop validation design, regulatory compliance (GDPR, HIPAA, FedRAMP), and ethical AI: bias, fairness, transparency. - [Stakeholder communication and lifecycle management — CCAR-P domain 06](https://thethirdwaycompany.com/guides/ccar-p/stakeholder-lifecycle): CCAR-P Domain 6 (14%), unique to the Professional tier: structured discovery and requirements, communicating architectural decisions and trade-offs, expectation alignment and honest SLAs, documentation and implementation guidance, and the discovery-to-iteration lifecycle. - [Developer productivity and operational enablement — CCAR-P domain 07](https://thethirdwaycompany.com/guides/ccar-p/developer-productivity): CCAR-P Domain 7 (7%): configuring Claude tools like Claude Code for teams, improving developer workflows with AI-assisted tooling, and supporting debugging and operational issue resolution. - [Claude Code, from zero to daily driver — the Beginner course](https://thethirdwaycompany.com/guides/claude-code): installment 1 of the Claude Code course — install and sign in, permission modes and plan mode, one change from request to commit, keeping a session on track, project memory with CLAUDE.md; 5 lessons, a capstone, 3 electives; adapted from Wesam Abousaid's MIT-licensed curriculum - [Install, sign in, and look around — Claude Code Beginner lesson 1](https://thethirdwaycompany.com/guides/claude-code/install): Claude Code Beginner lesson 1: install, first sign-in, workspace trust, and the three things to check about any session — version, model, account. - [Permission modes and plan mode — Claude Code Beginner lesson 2](https://thethirdwaycompany.com/guides/claude-code/permission-modes): Claude Code Beginner lesson 2: the four permission modes, Shift+Tab, and plan mode — see the whole change as a plan before a single file moves. - [Your first change, from request to commit — Claude Code Beginner lesson 3](https://thethirdwaycompany.com/guides/claude-code/first-change): Claude Code Beginner lesson 3: explore, plan, implement, test, review the diff line by line, correct, commit — the full loop, once, by hand. - [Keep a session on track — Claude Code Beginner lesson 4](https://thethirdwaycompany.com/guides/claude-code/session-control): Claude Code Beginner lesson 4: /context, /compact, /clear, /rewind and /resume. Long sessions drift; these five commands are the wheel and the pedals. - [Project memory with CLAUDE.md — Claude Code Beginner lesson 5](https://thethirdwaycompany.com/guides/claude-code/project-memory): Claude Code Beginner lesson 5: CLAUDE.md, /init and /memory. What every session should know about the project, written once, read at the start of every session. - [Capstone: fix a reported bug, from plan to commit — Claude Code Beginner capstone](https://thethirdwaycompany.com/guides/claude-code/capstone): The Claude Code Beginner capstone: a real reported bug, taken from plan to failing test to fix to commit — every habit from the five lessons in one 30–60 minute task. - [Elective: Claude Code in your editor — Claude Code Beginner elective](https://thethirdwaycompany.com/guides/claude-code/ide-extensions): Claude Code Beginner elective: the VS Code extension and the JetBrains plugin — reference the lines you are looking at, review edits in the editor’s diff view. - [Elective: give Claude a screenshot — Claude Code Beginner elective](https://thethirdwaycompany.com/guides/claude-code/screenshots): Claude Code Beginner elective: images in — drag, paste, or point at a path. Errors, layouts and mockups: some things are faster to show than to describe. - [Elective: learn from Claude Code itself — Claude Code Beginner elective](https://thethirdwaycompany.com/guides/claude-code/built-in-teachers): Claude Code Beginner elective: /powerup, the Learning and Explanatory output styles, and the claude-code-guide subagent — the tool teaches its own features. - [Claude Code, shaped to your project — the Intermediate course](https://thethirdwaycompany.com/guides/claude-code/intermediate): installment 2 of the Claude Code course — skills, project memory with imports and scoped rules, model and effort selection, permission scopes and the sandbox, hooks that enforce, subagents, MCP servers, plugins; 8 lessons and a capstone; adapted from Wesam Abousaid's MIT-licensed curriculum - [Turn a repeated workflow into a skill — Claude Code Intermediate lesson 1](https://thethirdwaycompany.com/guides/claude-code/intermediate/first-skill): Claude Code Intermediate lesson 1: write a project skill in .claude/skills/ that runs when you type its name and loads on its own when a request matches its description. - [Organize project memory, and see what loaded — Claude Code Intermediate lesson 2](https://thethirdwaycompany.com/guides/claude-code/intermediate/organize-memory): Claude Code Intermediate lesson 2: split CLAUDE.md into @imports and path-scoped rules in .claude/rules/, and find out which instruction files actually loaded when one is ignored. - [Pick the model and the effort — Claude Code Intermediate lesson 3](https://thethirdwaycompany.com/guides/claude-code/intermediate/model-and-effort): Claude Code Intermediate lesson 3: two knobs — the model decides how capable the answer can be, the effort decides how hard Claude works per request. /model, /effort, ultrathink, and frontmatter. - [Permissions, settings scopes, and the sandbox — Claude Code Intermediate lesson 4](https://thethirdwaycompany.com/guides/claude-code/intermediate/permissions-and-sandbox): Claude Code Intermediate lesson 4: the four settings files and who each affects, permission rules that commit with the repo, and the OS-enforced sandbox that bounds what shell commands can touch. - [Enforce a rule with a hook — Claude Code Intermediate lesson 5](https://thethirdwaycompany.com/guides/claude-code/intermediate/hooks): Claude Code Intermediate lesson 5: a PreToolUse hook that blocks edits to protected files with exit code 2, a PostToolUse hook that tidies up after — rules that hold in every permission mode. - [Delegate to a custom subagent — Claude Code Intermediate lesson 6](https://thethirdwaycompany.com/guides/claude-code/intermediate/subagents): Claude Code Intermediate lesson 6: a subagent in .claude/agents/ with only the tools its task needs — its file reads stay out of your context window, and only its summary comes back. - [Connect a tool with MCP — Claude Code Intermediate lesson 7](https://thethirdwaycompany.com/guides/claude-code/intermediate/mcp): Claude Code Intermediate lesson 7: MCP scopes — local for you, .mcp.json committed for everyone — and one server, Chrome DevTools MCP, added both ways with its version pinned and its data flows off. - [Install and manage plugins — Claude Code Intermediate lesson 8](https://thethirdwaycompany.com/guides/claude-code/intermediate/plugins): Claude Code Intermediate lesson 8: a plugin packages skills, hooks, subagents and MCP servers into one installable unit — marketplaces, scopes, what an enabled plugin costs every session, and the teammate step. - [Capstone: get a repository ready for a new teammate — Claude Code Intermediate capstone](https://thethirdwaycompany.com/guides/claude-code/intermediate/capstone): The Claude Code Intermediate capstone: a skill, scoped memory, team permissions and the sandbox, a protective hook, a read-only reviewer, a pinned MCP server and a plugin — one repo a new teammate can clone and use on day one. - [Claude Code, beyond one developer — the Advanced course](https://thethirdwaycompany.com/guides/claude-code/advanced): installment 3 of the Claude Code course — parallel sessions with worktrees, orchestration patterns, dynamic workflows, headless scripting with claude -p, GitHub Actions, sharing a team setup, bounded autonomous runs; 7 lessons and a capstone; adapted from Wesam Abousaid's MIT-licensed curriculum - [Parallel sessions with worktrees — Claude Code Advanced lesson 1](https://thethirdwaycompany.com/guides/claude-code/advanced/parallel-sessions): Claude Code Advanced lesson 1: run two sessions at once, each in its own git worktree, and merge both branches — isolation, what worktrees share, and cleanup. - [Choose an orchestration pattern — Claude Code Advanced lesson 2](https://thethirdwaycompany.com/guides/claude-code/advanced/orchestration-patterns): Claude Code Advanced lesson 2: main conversation, subagents, dynamic workflows or agent teams — three questions in order pick the cheapest pattern, and /tasks and /usage show the bill. - [Run and save a dynamic workflow — Claude Code Advanced lesson 3](https://thethirdwaycompany.com/guides/claude-code/advanced/dynamic-workflows): Claude Code Advanced lesson 3: have Claude write a workflow script that fans out across many files, keep the run small with a size guideline, and save it as a command. - [Script Claude Code with claude -p — Claude Code Advanced lesson 4](https://thethirdwaycompany.com/guides/claude-code/advanced/headless): Claude Code Advanced lesson 4: run one prompt with no one at the keyboard, branch on the JSON result and exit code, and cap the run with dontAsk mode, allow rules and turn limits. - [Claude Code in GitHub Actions — Claude Code Advanced lesson 5](https://thethirdwaycompany.com/guides/claude-code/advanced/github-actions): Claude Code Advanced lesson 5: run Claude unattended in a GitHub Actions workflow — the credential in a repository secret, the limits written in the file, and the teardown that retires it. - [Share your setup with a team — Claude Code Advanced lesson 6](https://thethirdwaycompany.com/guides/claude-code/advanced/share-your-setup): Claude Code Advanced lesson 6: commit the settings everyone should get, keep personal files out of git, and ship a team plugin from a marketplace inside the repository — no installs. - [Put bounds on autonomous runs — Claude Code Advanced lesson 7](https://thethirdwaycompany.com/guides/claude-code/advanced/bounded-runs): Claude Code Advanced lesson 7: bound an unattended run with least privilege, the strict sandbox, turn and spend caps, a review gate and a cost estimate from a slice run. - [Capstone: CI review your team can share — Claude Code Advanced capstone](https://thethirdwaycompany.com/guides/claude-code/advanced/capstone): The Claude Code Advanced capstone: two parallel worktree sessions, a bounded review script, a team plugin, a saved workflow and a pull-request review in CI — in a copy you tear down. ## Free tools - [Automation ROI calculator](https://thethirdwaycompany.com/tools): what repetitive work costs and what automation could recover - [Agent-readiness check](https://thethirdwaycompany.com/tools/agent-ready): score any public URL on the conventions AI agents rely on — the same 15 scored checks this site passes, plus detection-only agentic-commerce signals (product markup, declared feeds, ACP/UCP files) - [Team AI skills-gap checker](https://thethirdwaycompany.com/tools/skills-gap): rate each role's AI use against its task mix — per-role gap scores, training priority order, and first exercises, all client-side - [Shadow-AI exposure check](https://thethirdwaycompany.com/tools/shadow-ai-check): ten questions on the AI tools employees use, the data flowing through them, and the controls around it — indicative 0-100 exposure score and the three areas to investigate first, all client-side - [AI usage policy generator](https://thethirdwaycompany.com/tools/ai-policy-generator): answer eight questions on team, data, tools, red lines, and review — get a complete employee AI acceptable-use policy in markdown to copy or download. Pure client-side template assembly, nothing generated - [AI savings & ROI calculator](https://thethirdwaycompany.com/tools/ai-savings-calculator): build the team role by role — hours on automatable work, loaded hourly cost — and get a conservative-to-aggressive monthly/yearly savings range, a payback window against an engagement cost band, and a printable/emailable summary. Pure client-side arithmetic, nothing generated - [AI Work Mix scorecard](https://thethirdwaycompany.com/tools/ai-work-mix): twelve questions across five third-way dimensions — how much hands-on work AI does, the human review loop, decision rights, tool sprawl, skills investment — an instant 0-100 work-mix score, the shape of the mix (over-delegated, under-delegated, side-by-side, balanced), and the two moves that matter first, all client-side ## For agents - Machine-readable API catalog: https://thethirdwaycompany.com/.well-known/api-catalog (RFC 9727), human docs at https://thethirdwaycompany.com/docs - Agent permissions and conventions: https://thethirdwaycompany.com/agents.md - Authentication: https://thethirdwaycompany.com/auth.md (nothing to authenticate to — all APIs are public read-only)