GitHub’s chief executive, Thomas Dohmke, has publicly backed a controversial internal Microsoft memo that urged managers to reflect on employees’ use of internal AI tools as part of performance conversations — calling the practice “totally fair game” and framing AI fluency as part of company culture rather than a narrow productivity metric.
In mid‑2025, an internal message circulated inside Microsoft from Julia Liuson — president of a division that oversees developer tools including GitHub Copilot — asking managers to include AI usage and learning as part of their “holistic reflections” on employee performance. The memo argued that “AI is now a fundamental part of how we work” and equated AI literacy with other core workplace competencies such as collaboration and data‑driven thinking.
The memo quickly became a public flashpoint because it linked AI adoption to performance discussions at a moment when Microsoft and its peers are aggressively embedding AI into product roadmaps and operations. That broader corporate push toward AI comes amid large restructuring and a re‑prioritization of skills; reporting around the same period highlighted deep workforce changes and a major shift in internal expectations for AI competence.
On a widely distributed technology podcast, GitHub CEO Thomas Dohmke defended the memo’s intent and nuance. He said managers should ask employees to reflect on whether they used tools such as GitHub Copilot, Microsoft Copilot, or Teams Copilot and what they learned — not to count “lines of code produced by AI” — and argued that using the company’s own platform is non‑negotiable for GitHub employees. Dohmke framed the conversation as aligning with a growth‑oriented culture that values learning and practical application of new tools.
There is also a reputational vector to consider. Public controversy over internal policies — particularly in high‑visibility firms — can spill into recruiting, customer perceptions, and regulatory attention. Executives pushing adoption must therefore balance speed with safeguards.
The debate triggered by the memo and Dohmke’s comments is an early test case of how major tech firms will socialize rapid, product‑level change inside large organizations. The tone of that socialization — whether it leans supportive, coercive, or punitive — will shape employee retention, product quality, and legal risk for years to come.
Source: Business Insider GitHub CEO says Microsoft's memo about evaluating AI use is 'totally fair game'
In mid‑2025, an internal message circulated inside Microsoft from Julia Liuson — president of a division that oversees developer tools including GitHub Copilot — asking managers to include AI usage and learning as part of their “holistic reflections” on employee performance. The memo argued that “AI is now a fundamental part of how we work” and equated AI literacy with other core workplace competencies such as collaboration and data‑driven thinking.The memo quickly became a public flashpoint because it linked AI adoption to performance discussions at a moment when Microsoft and its peers are aggressively embedding AI into product roadmaps and operations. That broader corporate push toward AI comes amid large restructuring and a re‑prioritization of skills; reporting around the same period highlighted deep workforce changes and a major shift in internal expectations for AI competence.
On a widely distributed technology podcast, GitHub CEO Thomas Dohmke defended the memo’s intent and nuance. He said managers should ask employees to reflect on whether they used tools such as GitHub Copilot, Microsoft Copilot, or Teams Copilot and what they learned — not to count “lines of code produced by AI” — and argued that using the company’s own platform is non‑negotiable for GitHub employees. Dohmke framed the conversation as aligning with a growth‑oriented culture that values learning and practical application of new tools.
Why this matters: business strategy meets workplace culture
Microsoft’s strategic context
Microsoft has made AI a strategic priority across products and services. Internally, leadership has signaled aggressive investments in tooling and infrastructure — and set expectations that employees move from curiosity to day‑to‑day use. Those signals create strong incentives for managers to push adoption within their teams. Multiple internal communiqués and reporting from forum summaries and news snippets show this has been a company‑wide message through 2025.GitHub’s stake
GitHub sits at the intersection of coding workflows and AI assistance. GitHub Copilot is both a product and a lever for encouraging modern software practices. Dohmke’s statement that there is “no world where I would allow somebody to say, ‘I don’t want to use GitHub’” signals a push to normalize the platform among all functions — not only engineering but also HR, sales, and legal — as part of the company’s operating norms. That emphasis doubles as product‑led vendor adoption: employees using company products internally drives learning, refinement, and evangelism.What the memo actually asks managers to do
- Treat AI usage and learning as part of a holistic evaluation, similar to collaboration or communication.
- Encourage employees to reflect on how they used AI (e.g., Copilot or Teams Copilot) and what they learned from those interactions.
- Avoid simple, gamable metrics (for example, a raw count of lines or characters generated by AI); focus instead on mindset, outcomes, and demonstrated learning.
The upside: why leaders argue this is sensible
- Alignment between strategic priority and employee capability. If an organization’s product roadmap relies on AI, expecting staff to understand and use AI tools is a logical extension of job requirements. When product and operations use the same stack, synergies emerge faster.
- Faster learning loops and better product feedback. Internal use of Copilot‑style tools helps surface real‑world problems, produces feedback loops for improvement, and accelerates de‑risking before broader customer rollouts.
- Productivity gains when used properly. Firms experimenting with AI in engineering and knowledge work report real reductions in repetitive toil and faster onboarding for junior contributors — savings that can improve throughput and job satisfaction when balanced against appropriate oversight.
- Cultural modernization. Firms that adopt emergent tools across roles can reposition their workforce for a competitive market that increasingly rewards AI literacy.
The real risks and trade‑offs managers must wrestle with
1) Surveillance, trust and psychological safety
Asking managers to evaluate employees on AI use can easily feel like surveillance. Even when framed as “learning,” employees may fear that reduced usage equals poor performance or a signal their role is at risk of automation. That psychological dynamic can poison team trust and increase stress — especially amid industry‑wide layoffs and reorganization. Multiple summaries of the broader environment in 2025 show employees interpreting AI pushes through the lens of job security.2) Gamable metrics and perverse incentives
Concrete counts (tokens used, lines generated, number of Copilot sessions) are trivial to game and have low correlation with actual value delivered. Measuring use instead of outcomes may reward activity rather than impact. Dohmke explicitly cautioned against simplistic metrics and endorsed a mindset evaluation; still, companies often drift into measurable but misleading KPIs.3) Privacy, IP, and compliance exposure
Widespread Copilot usage outside tightly controlled environments raises data governance questions. Depending on configuration, AI assistants can surface proprietary code snippets, reveal private information, or call external models in ways that violate confidentiality or compliance rules. Any policy tying AI use to performance must be coupled with clear, enforceable data‑handling rules and tool configurations that protect sensitive assets. Forum reporting and enterprise‑grade analyses repeatedly flag these risks for teams adopting Copilot and similar assistants.4) Equity and accommodation
Not every employee starts from the same technical baseline. Non‑technical roles, older employees, or those with limited access to training may be disadvantaged if AI fluency becomes a de facto gate to advancement. Organizations must guard against inadvertently penalizing those who need time, resources, or accommodations to upskill.5) Legal and labor implications
Performance criteria must meet legal standards of fairness and non‑discrimination. If AI usage metrics intersect with protected characteristics (age, disability, religious practices affecting work patterns), companies may face legal exposure or labor disputes. In some cases, these policies have already drawn internal pushback and public scrutiny.How to measure AI readiness and contribution fairly (practical framework)
Measuring AI usage as part of performance conversations can be defensible if it’s outcome‑focused, transparent, and supportive. The following framework is intended for managers and HR teams designing fair implementation:What managers should measure (qualitatively and quantitatively)
- Evidence of learning and intent — documented reflections on what the employee tried with AI, what they learned, and how they applied it.
- Clear outcomes — did AI use reduce cycle time, improve code quality, or produce better client deliverables? Concrete examples matter more than raw counts.
- Peer and manager observations — did use of AI enable better collaboration, faster reviews, or clearer documentation? Third‑party observations reduce self‑reporting bias.
- Security and compliance adherence — did the employee follow company rules for data handling while using AI tools?
- Complexity of tasks — did AI support high‑value tasks or mostly trivial chores? Value density matters.
What managers should not measure
- Number of Copilot invocations, raw lines of code generated by AI, or session counts without context. These are easily gamed and low signal.
Implementation steps (1. to 6.)
- Define desired outcomes and examples of acceptable AI use for each role.
- Build voluntary, role‑specific learning paths (sessions, micro‑certifications, internal docs).
- Train managers on evaluating learning artifacts and outcome evidence — not activity logs.
- Create opt‑in pilot programs before broad rollouts to refine language and guardrails.
- Log and surface compliance issues separately from performance; breaches should be handled with HR and legal, not conflated with adoption conversations.
- Offer appeal and coaching pathways for employees flagged for low AI engagement.
A balanced HR/legal checklist for rolling this out
- Update job descriptions to explicitly list AI competency expectations where relevant.
- Publish clear policies on data classification, permitted prompts, and tool configuration.
- Ensure training is timely, role‑appropriate, and resourced.
- Create an appeal process and human‑review step before any negative performance outcome is recorded against AI use.
- Monitor for disparate impact across demographics and supply accommodations.
- Involve legal counsel early to align performance criteria with employment law and collective bargaining agreements where applicable.
What employees can do to protect their interests
- Keep a short log of AI experiments and outcomes — what prompt you used, what the AI produced, and how you validated or edited that output. That documentation turns vague “did you use AI?” questions into demonstrable learning artifacts.
- If worried about IP or privacy, rely on approved tool configurations and consult the security team before pasting sensitive code or documents into AI assistants.
- Push for training, time, and clarity from managers; decline to be judged on opaque or unfair metrics.
- If metrics appear gamed or misapplied, escalate through HR or employee representative channels.
Broader implications: workforce strategy and the future of craft
The debate over AI in performance reviews is a microcosm of a larger tension: how companies manage technological disruption without eroding worker autonomy or trust. When a platform owner (GitHub) also produces a productivity tool (Copilot), internal adoption goals can be both product strategy and workforce strategy. The challenge is to harness the upside — productivity, faster learning, product improvement — without sliding into a culture of coercion or surveillance.There is also a reputational vector to consider. Public controversy over internal policies — particularly in high‑visibility firms — can spill into recruiting, customer perceptions, and regulatory attention. Executives pushing adoption must therefore balance speed with safeguards.
Critical assessment: what GitHub and Microsoft are doing well — and where they still need to prove it
Notable strengths
- Leadership alignment: top executives are clear that AI is strategic, which reduces ambiguity in resourcing and product focus.
- Emphasis on learning: messaging from leaders emphasizes reflection and learning, not pure surveillance — a critical distinction that can be operationalized.
- Rapid iteration: internal dogfooding and aggressive deployment can fast‑track product improvement and real‑world safety checks.
Open questions and risks that remain
- Implementation fidelity: leaders can say “focus on learning,” but many organizations default to measurable proxies that are misleading. Without training and guardrails, the memo risks producing the wrong incentives.
- Worker trust: past layoffs and a faster pivot to AI have sensitized the workforce; a memo that ties AI use to performance will be interpreted in that context. Forum reporting from the period shows employees already concerned about automation and job security.
- Data governance: Copilot and similar tools require careful configuration to prevent accidental leakage of sensitive code or data; any performance program must be coupled to strict compliance measures.
Final takeaways: how to make AI‑aware performance reviews fair and constructive
- Treat AI competence as a learning objective, not a surveillance metric. Emphasize reflection, outcomes, and documented learning.
- Avoid raw usage counts; instead, evaluate the quality of AI application and whether it produced valuable outcomes.
- Invest meaningfully in training, accommodations, and clear security guardrails before making adoption a formal evaluation axis.
- Build transparent HR processes with appeals and human review to prevent unfair penalization.
- Monitor for unintended consequences across diversity and inclusion metrics and adjust policies when disparities appear.
The debate triggered by the memo and Dohmke’s comments is an early test case of how major tech firms will socialize rapid, product‑level change inside large organizations. The tone of that socialization — whether it leans supportive, coercive, or punitive — will shape employee retention, product quality, and legal risk for years to come.
Source: Business Insider GitHub CEO says Microsoft's memo about evaluating AI use is 'totally fair game'