How we use AI
We use AI in our own work every day.
Most consulting practices tell you what to do with AI without showing what they do with it. This page is the other way around: named tools, named governance decisions. The same methodology, frameworks, and measurement standards we apply to client engagements govern our own AI use. If we wouldn't recommend it to a client, we don't run it ourselves.
Why we publish this: research from Duke (PNAS 2025, n>4,400) found people who use AI at work are judged as lazier and less competent, except by evaluators who also use AI.12 The penalty disappears when use is visible. So we show.
Where we operate
We operate at Coach level.
The same four cognitive levels we train clients through (Intern, Coworker, Teacher, Coach) apply internally. We operate at Coach level: AI challenges assumptions, stress-tests decisions, surfaces blind spots, and pushes back on weak reasoning. Every public artifact on this site, every client deliverable, and every engagement design has been tested against AI as a thinking partner before it ships. The human owns the output. The AI earns its seat by making the output harder to get wrong.
Tools we use
AI tools running in our daily practice, for drafting, code, research, and operations. Tool-specific knowledge decays in roughly 90 days, so the list below is dated. Check the footer's last-reviewed stamp before assuming any of it is still current.
Claude Code (Anthropic)
Primary surface for code, scripts, and technical documentation. Far West runs a production-scale Claude Code environment with multi-agent workflows and MCP integrations, the same infrastructure proposed to clients.
Claude Pro + Projects (Anthropic)
Primary surface for client-facing drafting: case studies, training-pack copy, internal memos, engagement designs. Engagement-specific Projects carry the customer's voice descriptors as instructions.
MCP integrations (multiple)
Model Context Protocol servers connecting Claude to working tools: Obsidian for the project knowledge base, Firecrawl for source-checking research claims, Sequential Thinking for multi-constraint planning, scheduled-tasks for recurring operations.
Microsoft Copilot for M365 (Excel, Outlook, Word)
For work that lives inside the suite where the artifact already is. Variance recaps in Excel, draft-then-edit in Outlook, structured memos in Word. We use it because clients use it.
ChatGPT, Gemini, Perplexity, NotebookLM
Cross-platform testing, research validation, and fluency across the tools we train clients on. Every tool in the training catalog is a tool we use, so instruction reflects current capability, not last quarter's documentation.
How we select tools for the work
The tool follows the workflow. Every time.
Structured drafting + complex reasoning
Claude
Code, scripts, technical builds
Claude Code
Microsoft-native workflows
Copilot
Research + source verification
Perplexity (speed), NotebookLM (depth)
Cross-platform fluency
ChatGPT + Gemini
Multi-tool orchestration
MCP integrations
The selection principle: No tool is default for everything. Each earns its place by fitting the workflow it serves. When a client asks "which AI tool should we use?" the answer is always "for which workflow?", and we can demonstrate the reasoning because we make the same decision daily.
How we govern our own AI use
The same frameworks we apply to clients govern this practice.
The governance standards we propose to clients (NIST AI RMF for risk identification and ISO/IEC 42001 for the management system layer) are applied to Far West's own AI infrastructure. Not a marketing claim. The operating standard.
Risk identification (NIST AI RMF, applied internally)
Every AI-assisted output carries risk: factual error, hallucination, voice drift, confidentiality exposure. We categorize risks by artifact type. Client-facing deliverables carry the highest scrutiny; internal operational use, standard controls; research and ideation, baseline verification.
Control design
Every cited source is checked against the primary before publication, the same vendor-claim discount rule applied to client engagements. AI-drafted content is reviewed line-by-line by the practitioner before it reaches any audience outside Far West.
Operational accountability
Every AI-assisted decision has an owner-of-record: Paul. No automated output ships to a client without human review. No AI-generated recommendation overrides practitioner judgment. The diagnostic engine recommends; the practitioner decides.
Data handling
Client data entering any AI tool follows our confidentiality and data governance standards. Engagement-specific data stays within engagement-specific Projects. No client data is used to train external models. NDAs are standard.
How we use AI for research
AI assists research. It doesn't replace verification.
AI tools accelerate literature review, source identification, and citation checking. They do not determine what the evidence says. Every claim on this site that cites a peer-reviewed study has been verified against the primary source, not against AI's summary of it. Where vendor productivity claims and peer-reviewed findings disagree, we default to the peer-reviewed evidence. The same standard applied to client engagements: design for the research, not the marketing.
What we don't outsource to AI
The interesting question isn't where AI shows up. It's where it doesn't. Three categories stay human in our practice, and they'd be the same three categories in yours.
Judgment calls.
Whether a client is the right fit for an engagement. Whether a recommendation is structurally honest or just lucrative. Whether a draft reads true or just clean. These decisions sit with the practitioner, not the tool.
Regulated decisions.
Anything touching client data with regulatory weight (PIPEDA, GDPR, sector-specific rules) goes through our governance frame, not through a generative model. AI assists drafting a memo about the rule; the rule itself is human-read, human-checked, human-cited.
Voice.
The voice in case studies, proposals, and on this page is Paul's. AI assists drafting and structure; the voice itself is checked sentence by sentence on every public artifact. If a sentence feels generative, it gets rewritten or cut.
The connection to client work
We run what we sell.
Every tool in the training catalog is a tool we use. Every governance framework we propose to clients governs our own operations. Every measurement standard we apply to client outcomes (behaviour change, output quality, workflow retention) is applied to our own AI use. The methodology isn't theoretical. The infrastructure isn't a demo. The four cognitive levels aren't a training model we describe. They're the operating levels we work at daily. If we wouldn't pass our own diagnostic, we wouldn't propose the diagnostic to clients.
References
12Reif, J.A., Larrick, R.P., & Soll, J.B. (2025). Evidence of a social evaluation penalty for using AI. Proceedings of the National Academy of Sciences. n>4,400 across four experiments. DOI: 10.1073/pnas.2426766122.
For the full evidence base referenced across Far West Consulting, see Sources.