--- entity: Frimer-Rasmussen Consulting title: Get - and Keep - Control Over Your Organization's Tacit Knowledge subtitle: AI is moving into the places where work gets discussed, judged, corrected, and approved. Leaders need to govern what it learns there. author: Mikkel Frimer-Rasmussen last_updated: '2026-06-25' type: TechArticle tech_stack: - Enterprise AI - AI Governance - Organizational Knowledge - AI Sovereignty status: Published primary_goal: Explain why shared AI agents make tacit organizational knowledge a strategic governance, portability, and continuity issue. canonical_url: https://frimer-rasmussen.dk/articles/get-and-keep-control-over-your-organizations-tacit-knowledge.html about: - name: Tacit knowledge url: https://en.wikipedia.org/wiki/Tacit_knowledge - name: Artificial intelligence url: https://en.wikipedia.org/wiki/Artificial_intelligence - name: Data Act url: https://digital-strategy.ec.europa.eu/en/policies/data-act mentions: - name: Anthropic url: https://www.anthropic.com/ - name: Claude url: https://www.anthropic.com/claude - name: European Commission url: https://commission.europa.eu/ concepts: - name: Workstream AI description: AI agents embedded in shared channels, repositories, queues, boards, and decision flows where organizational work actually happens. - name: Portable AI Workflow description: A documented, versioned workflow that captures role, sources, criteria, examples, approvals, logs, and fallback paths outside one vendor tool. - name: Tacit Knowledge Dependency description: The operational risk that a single AI platform absorbs enough informal organizational memory to become difficult to replace. --- ## AI is moving into the places where work gets discussed, judged, corrected, and approved. Leaders need to govern what it learns there. Get - and keep - control over your organization's tacit knowledge. That is the management task behind the next phase of enterprise AI. Your company runs on more than documents, dashboards, and process maps. It runs on the exception someone remembers from last year's customer crisis. The product compromise that never reached the roadmap. The support thread where the team learned how a key account actually thinks. The review comment that defines "good enough" better than the official quality manual. This knowledge lives in the work. Now AI is moving into the work. On June 24, 2026, Economic Times reported that Anthropic had launched Claude Tag, a Slack-native AI experience for Team and Enterprise customers. According to the report, Claude can be tagged in Slack channels, work with approved organizational tools and code repositories, remember relevant channel and workspace context over time, continue tasks asynchronously, notify users when work is done, and operate under administrative controls such as channel scoping, activity logs, and spend limits. Claude Tag may win. Another product may win. The market will decide. The signal already matters: AI agents are entering the shared workstream. They will read the discussion before the decision. They will see disagreement before alignment. They will learn the informal rules behind the formal process. They will help teams move faster, and that help will teach them how the organization thinks. Treat that as a leadership issue from day one. ## The Workstream Becomes The Interface The first wave of generative AI felt personal. One employee opened a chat window, asked a question, drafted a text, summarized a document, wrote code, or analyzed a spreadsheet. Companies bought seats. IT wrote guidelines. Managers encouraged experimentation. That phase created real value. The next phase changes the unit of adoption. AI agents will sit in Slack, Teams, code repositories, support queues, product boards, research folders, customer operations, finance processes, and internal decision flows. They will follow work across days and weeks. They will connect fragments that people usually carry in memory. They will push tasks forward while teams handle other priorities. This shifts the executive question. A CEO must ask how AI changes the operating model. A CIO must ask how access, logging, security, integration, and vendor risk change when AI joins live workspaces. An innovation leader must ask which pilots can become durable organizational capabilities. A product development leader must ask how AI will absorb customer complaints, roadmap logic, technical constraints, bug triage, and product quality standards. The same organizational context creates the value and the dependency. ## Tacit Knowledge Creates The Deepest Dependency Most leadership teams know classic software lock-in. They know the pain of changing CRM, ERP, cloud, analytics, or productivity platforms. They understand migration projects, retraining, integrations, contracts, and data exports. AI adds a more subtle dependency. An agent learns patterns. It learns which sources people trust. It learns which arguments persuade leadership. It learns how engineers evaluate risk. It learns how commercial teams talk about customers. It learns which quality checks matter when the deadline is real. It learns the organization's shortcuts, escalation paths, and hesitation points. Run enough important work through one AI platform, and the platform starts to hold operating memory around the work. The company may still own the documents. It may still export the raw data. It may still pass a procurement checklist. The practical capability can still become hard to reproduce. Could another model run the same workflow with the same quality? Could another provider understand the same context? Could the team explain why the AI made a recommendation? Could the process continue if access changed next week? Put those questions in the boardroom and in the AI policy. ## Make The Workflow Movable The EU Data Act gives this discussion an important legal backdrop. The European Commission describes the Data Act as a framework that gives users more control over data and supports switching between providers of data-processing services. It has applied since September 12, 2025. That progress helps companies move data. Enterprise AI requires another layer: the ability to move work. A mature AI workflow contains a full operating pattern: - the role the AI may play; - the sources it may use; - the decisions it may support; - the criteria it must apply; - the examples that define good work; - the edge cases it must handle; - the assumptions it must expose; - the approvals humans must give; - the logs auditors may need; - the fallback path when the system fails. This is intellectual capital. Treat it like an asset. Store it in your own formats. Version it. Test it. Review it. Improve it. Make it movable. Companies that do this will gain from AI while keeping control of the practical knowledge that makes them good. ## Sovereignty Becomes A Continuity Question Many executives hear "AI sovereignty" and think about geopolitics, regulation, or national infrastructure. Inside a company, the issue quickly becomes operational: Can we keep working? Can product development continue if a model changes behavior? Can customer operations continue if a vendor changes access? Can a compliance workflow continue if a customer demands stricter data handling? Can a finance process continue if token costs rise? Can a quality review continue if a supplier removes the feature your team built around? On June 24, 2026, Business Insider reported that a legal tech startup sued the U.S. government after restrictions affected access to Anthropic's advanced models. The company claimed the loss of access caused immediate and severe operational harm. The case may develop in several directions, but the management lesson already stands: external model access creates availability risk when critical workflows depend on it. This should make companies more disciplined and more ambitious. Use American AI systems when they deliver the best value. Use European systems when the risk profile demands it. Use open models where control matters. Use more than one supplier when resilience justifies the cost. Design the dependency consciously. ## Budget AI As Delegated Work Shared AI agents also change the economics. Traditional software budgets revolve around seats, licenses, subscriptions, and modules. AI agents spend while they work. They can read more material, call tools, retry, summarize, investigate, generate, evaluate, and escalate. A team can create value through that activity. It can also burn budget while producing noise. Cost governance needs to follow the work. Ask: - Which workflows may use AI autonomously? - Which tools may the agent call? - Which spend limits apply per channel, process, customer, or risk category? - Which outputs require human approval? - Which metrics prove value? - Who owns the budget when AI work crosses departments? Seat price tells only a small part of the story. Leaders need to know how much delegated work the company buys, under which controls, and with which documented effect. ## Six Actions Leaders Should Take Now Companies should use shared AI agents. The productivity upside is real. The organizational memory upside may become even larger. Strong governance makes that ambition safer. Use this rule: Every critical AI workflow must have a portable version. Then build six controls. First, write down the workflow outside the vendor tool. Store prompts, system instructions, role definitions, source lists, examples, decision rules, quality criteria, and approval steps in the company's own repository or document management system. Second, version the quality standard. When AI helps produce product requirements, risk reviews, customer responses, supplier assessments, financial analysis, legal drafts, or technical specifications, record the criteria used and the date they changed. Third, classify AI work by business criticality. A drafting assistant does not need the same controls as an AI-supported process that influences customer commitments, compliance-sensitive analysis, product priorities, or executive decisions. Fourth, test critical workflows on more than one model or provider. You do not need multi-model architecture everywhere. You do need evidence that critical work can move when the business requires it. Fifth, log the reasoning trail. Record the sources used, assumptions made, alternatives rejected, human interventions, approvals, and final decisions. Output alone gives too little accountability. Sixth, define the fallback before the outage. Customer-critical and compliance-critical processes need a continuation plan before a model, integration, feature, vendor, or access route fails. These controls make serious AI adoption possible. ## The Leadership Question Claude Tag gives leaders a preview of enterprise AI's next stage. AI will join shared channels, follow work, build context, and help teams complete tasks across systems. Companies that govern this well will move faster and remember more. Companies that treat it as another chat feature will discover the governance problem late. Ask the question now: Can we document, quality-assure, budget, govern, and move the AI workflows our business starts to depend on? A clear yes gives shared AI agents room to become a strategic capability. An unclear answer leaves the company exposed to a quiet dependency: renting access to its own collective intelligence. Get the productivity gain. Keep control of the tacit knowledge. That is the work. ## Sources - Economic Times, "Anthropic launches Claude Tag: Everything you need to know about the Slack-native AI agent", June 24, 2026: https://economictimes.indiatimes.com/tech/artificial-intelligence/anthropic-launches-claude-tag-everything-you-need-to-know-about-the-slack-native-ai-agent/articleshow/131958086.cms - European Commission, "Data Act", last updated December 15, 2025: https://digital-strategy.ec.europa.eu/en/policies/data-act - Business Insider, "An AI startup is suing the US government for taking away Anthropic's new model", June 24, 2026: https://www.businessinsider.com/legion-ai-startup-suing-us-government-new-anthropic-model-fable5-2026-6