Detailed Analysis
A small management consulting firm's internal debate over how to deploy Claude reflects a broader tension playing out across knowledge-work industries as they grapple with AI adoption strategy. The Reddit post frames the disagreement as a binary: pursue "tool implementation first," focused on Claude's native PowerPoint and Excel generation capabilities for immediate productivity gains, versus a "deeply integrated knowledge and reasoning management" approach using Claude Code and CLAUDE.md configuration files distributed throughout SharePoint to create a persistent, firm-wide knowledge and reasoning layer. This is not a trivial disagreement — it represents two fundamentally different bets about where value accrues from AI adoption in professional services.
The tool-first camp's approach leverages Anthropic's recent investments in Claude's file-creation capabilities, which allow the model to generate and edit native Office documents directly, a feature set Anthropic rolled out specifically to compete for enterprise workflows historically dominated by Microsoft Copilot. This path is attractive to a ten-person firm because it requires minimal setup, delivers visible time savings on deliverables like client decks and financial models, and lets consultants see ROI within days. Its weakness is that it treats Claude as a document-generation utility rather than a repository of institutional knowledge — each engagement effectively starts from scratch, with the model reasoning only from what's placed in its context window at that moment.
The CLAUDE.md/Claude Code approach is more ambitious and technically demanding: it treats the firm's SharePoint environment as a structured knowledge base, using markdown configuration files to encode firm methodology, past engagement learnings, client context, and analytical frameworks that Claude can reference consistently across projects. This mirrors how software engineering teams use CLAUDE.md files to give Claude persistent project context, but repurposed for consulting deliverables and institutional memory rather than codebases. The payoff is a compounding asset — the more the firm invests in documenting its reasoning patterns and frameworks, the smarter and more consistent Claude's outputs become across the firm, reducing reliance on any single consultant's tacit knowledge. The cost is real: someone has to architect the file structure, maintain it as engagements evolve, and train non-technical staff to work with what is essentially a lightweight knowledge-management system built on developer tooling.
This debate is emblematic of a broader pattern across small and mid-sized professional services firms adopting frontier AI models in 2025-2026: the tension between quick-win productivity tooling and infrastructure investment that pays off over a longer horizon. Larger consulting firms like McKinsey, BCG, and Deloitte have the resources to run both tracks simultaneously — deploying consumer-facing AI tools for immediate use while separately building proprietary knowledge platforms on top of models like Claude via Anthropic's enterprise API. Smaller shops, by contrast, must choose where to place their limited implementation bandwidth, and the choice has real strategic stakes: firms that under-invest in structured knowledge management risk falling behind competitors who successfully turn Claude into a defensible, compounding institutional asset rather than just a faster word processor. The fact that this argument is happening inside a ten-person firm underscores how quickly capable agentic tools like Claude Code have democratized approaches to AI infrastructure that were, until recently, the province of much larger organizations with dedicated engineering teams.
Read original article →