Thomas Dohmke, the chief executive of GitHub, has forcefully defended a controversial internal Microsoft memo that urged managers to factor AI tool usage into employee performance evaluations, calling the practice "totally fair game" and framing AI fluency as a core cultural expectation rather than a narrow productivity metric. The endorsement, delivered during an August 7 appearance on the Decoder podcast, came just days before Dohmke surprised the industry by announcing his departure from the top role at GitHub at the end of the year.

The convergence of these events—a combative defense of AI-centric performance reviews and a high-profile executive exit—caps a tumultuous period inside Microsoft and its subsidiaries, where leadership is racing to embed artificial intelligence into every layer of operations while simultaneously reshaping workforce priorities through restructuring and redefined skill expectations.

The memo that sparked a firestorm

In mid-2025, Julia Liuson, the president of the Microsoft division responsible for developer tools including GitHub Copilot, circulated an internal message that immediately became a lightning rod for debate. The memo explicitly instructed managers to include employees' use of AI tools as part of their "holistic reflections" on performance, arguing that AI literacy had become as fundamental as collaboration or data-driven thinking.

"AI is now a fundamental part of how we work," the memo stated. "Just like collaboration, data-driven thinking, and effective communication, using AI is no longer optional—it's core to every role and every level."

The directive landed at a sensitive moment. Throughout 2025, Microsoft had been aggressively pushing AI across its product ecosystem—from Copilot in Microsoft 365 to Azure AI services—while simultaneously carrying out significant workforce reductions and internal reorganizations. The memo's linkage of AI adoption to performance conversations felt, to many employees, less like an invitation to learn and more like a ultimatum cloaked in corporate jargon.

Dohmke doubles down: 'Totally fair game'

Asked about the controversy on the Decoder podcast, Dohmke did not retreat. Instead, he defended the memo's intent and nuance, insisting that the conversation between manager and employee should center on reflection and learning, not on mechanistic counting.

"I think in 2025, it's totally fair game to say you should reflect on your AI usage, and you should reflect what did you learn about AI, did you use GitHub Copilot or Microsoft Copilot, Teams Copilot to summarize a meeting, and if not why not?" Dohmke said.

He explicitly warned against gamifying the process with shallow metrics. "Measuring AI usage doesn't mean assessing how many lines of code someone wrote with AI," he said, noting that such figures are "easily gamified." Instead, he framed AI competency as a mindset that aligns with Microsoft and GitHub's growth-oriented culture.

Dohmke's status as CEO of GitHub—the home of Copilot, one of the most widely adopted AI coding assistants—gave his words particular weight. As both a product leader and an internal customer, his insistence that AI tool use is non-negotiable reflects a dual strategy: driving product-led adoption and ensuring that every employee becomes a fluent user of the company's own technology.

"There is no world where I would allow for somebody to say, 'Well, sorry, I don't want to use GitHub,'" he said. "And I think that's fair game—if the employee doesn't want that, then there's tens of thousands of other tech companies out there where they can have that."

That expectation, he clarified, extends beyond engineering to every division: HR, sales, legal, and operations. At GitHub, using the platform is as mandatory as breathing air.

The broader Microsoft AI push: context and friction

Dohmke's comments cannot be separated from Microsoft's larger strategic ambitions. Under CEO Satya Nadella, the company has bet billions on AI infrastructure and product integration. Internal communications throughout 2025 have reinforced a company-wide mandate to move from AI curiosity to routine, daily use. This push is not merely aspirational; it is backed by real investment, product roadmaps, and organizational restructuring.

Multiple internal communiqués and reporting from the period show that leadership has been consistent: AI competence is now a baseline expectation. For Microsoft and its subsidiaries, internal adoption serves multiple purposes. It generates rapid feedback loops for product improvement, de-risks features before broader rollout, and turns the workforce into a living advertisement for the technology.

Yet this alignment of business strategy with employee evaluation has proven deeply contentious. Workers, already anxious from waves of tech layoffs and headcount reductions, interpreted the memo through a lens of job security. If AI usage becomes a performance metric, does lower usage signal a lack of commitment—or worse, a replaceable skill set?

What the memo actually asks managers to do

According to the guidance shared by Liuson and echoed by Dohmke, the expectations are meant to be qualitative and developmental, not punitive or surveillant. The key elements include:

  • Holistic evaluation: Treat AI usage as one component of overall performance, akin to communication or collaboration skills.
  • Reflection over counting: Prompt employees to discuss which AI tools they used, what they tried, and what they learned—not to report raw usage counts.
  • Outcome orientation: Focus on whether AI use improved efficiency, quality, or creativity, and to document those outcomes.
  • Cultural alignment: Position AI fluency as a natural extension of working for a company that builds AI tools.

In theory, this approach avoids the trap of reductionist metrics. In practice, implementation is far messier.

The real risks: surveillance, equity, and perverse incentives

The backlash to the memo is rooted in legitimate concerns that any performance-linked AI policy must confront.

Surveillance and psychological safety

Even when framed as "learning," any evaluation tied to tool usage can feel like monitoring. Employees may worry that reduced interaction with AI—whether due to task requirements, personal preference, or simply forgetting to log sessions—will be held against them. In a climate of ongoing layoffs, that fear is amplified. Trust is fragile; once broken, it poisons team dynamics and increases burnout.

Gamable metrics and misleading KPIs

Dohmke's warning against counting lines of code is well taken, but organizations often drift toward what is measurable rather than what is meaningful. The temptation to automate dashboards showing Copilot sessions, prompts submitted, or tokens generated is immense. These metrics, while easy to track, have almost no correlation with actual value or learning. They risk rewarding superficial engagement while penalizing deep, thoughtful use that is harder to quantify.

Privacy, IP, and compliance exposure

AI assistants like Copilot raise serious data governance questions. Depending on configuration, they can inadvertently surface proprietary code, leak confidential information, or call external models in ways that violate regulatory requirements. Any policy that ties performance to AI use must be coupled with ironclad data-handling rules, approved tool configurations, and explicit guardrails. Without these, the organization opens itself to severe reputational and legal risk.

Equity and accommodation

AI fluency is not equally distributed. Employees in non-technical roles, older workers, or those with limited prior exposure to AI tools may need significant time and training to upskill. If AI competence becomes a de facto gate to advancement, organizations risk systemic exclusion. Fair implementation demands clear pathways for learning, accommodations for varied starting points, and safeguards against biased outcomes.

Performance criteria tied to AI use must meet legal standards of fairness and non-discrimination. If such metrics intersect with protected characteristics—age, disability, or religious practices that affect work patterns—companies could face lawsuits or labor disputes. Internal pushback and public scrutiny are virtually guaranteed unless criteria are transparent, appealable, and applied with human judgment.

A practical framework for measuring AI readiness fairly

For organizations considering similar policies, the following framework—drawn from the forum discussion on workplace culture—offers a starting point for defensible implementation.

What to measure

  • Evidence of learning: Documented reflections on AI experiments, insights gained, and applications attempted.
  • Concrete outcomes: Cases where AI use reduced cycle time, improved quality, or enabled better deliverables. Real examples matter more than abstract claims.
  • Peer and manager observations: Third-party validation of improved collaboration, faster code reviews, or clearer documentation facilitated by AI.
  • Security and compliance adherence: Demonstrated adherence to company data-handling policies while using AI tools.
  • Task complexity: Whether AI supported high-value, complex tasks or was relegated to trivial chores.

What not to measure

  • Raw invocation counts, AI-generated lines of code, or session durations without context.
  • Any single numeric metric that can be trivially inflated.

Implementation steps

  1. Define role-specific desired outcomes and examples of acceptable AI use.
  2. Create voluntary, role-appropriate learning paths (micro-certifications, internal workshops).
  3. Train managers to evaluate learning artifacts and outcomes, not activity logs.
  4. Pilot the program with opt-in teams before wide rollout to refine guardrails.
  5. Separate compliance monitoring from performance assessment; handle breaches through HR and legal, not performance ratings.
  6. Provide appeal and coaching pathways for employees flagged for low AI engagement.
  • 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 adequately resourced.
  • Create an appeal process with human review before any negative performance outcome is recorded based on AI use.
  • Monitor for disparate impact across demographic groups and supply reasonable accommodations.
  • Involve legal counsel early to align criteria with employment laws and collective bargaining obligations.

For employees: protecting your interests

  • Keep a personal log of AI experiments and outcomes—what prompt you used, what the AI produced, and how you validated or edited it. This documentation turns vague "did you use AI?" inquiries into demonstrable evidence of learning.
  • Use only approved tool configurations; never paste sensitive code or documents into AI assistants without explicit security approval.
  • Demand training, time, and clarity from managers; refuse to be judged on opaque or unfair metrics.
  • Escalate through HR or employee representatives if metrics become gamed or misapplied.

The resignation bombshell

In a twist that no one saw coming, Dohmke announced on Monday—just four days after the podcast aired—that he will step down as GitHub's CEO at the end of the year. In a blog post, he said GitHub and its leadership team will remain part of Microsoft's CoreAI organization, the unit that oversees the company's most critical AI initiatives.

Dohmke did not explicitly link his departure to the performance review controversy. His statement focused on the next chapter for both himself and GitHub. Yet the timing inevitably invites questions. Was the AI performance debate a factor? Did internal friction over the aggressive AI push play a role? Or is this simply a planned transition after several years at the helm?

Whatever the reason, the departure adds another layer of uncertainty to a company navigating one of the most significant technological shifts in its history. Dohmke's successor will inherit not only a wildly successful platform but also the cultural tensions that come with enforcing AI adoption from the top down.

Critical assessment: strengths and open questions

What Microsoft and GitHub are doing right

  • Leadership alignment: The clear, consistent message from top executives reduces ambiguity and ensures resourcing for AI initiatives.
  • Emphasis on learning: Dohmke's repeated focus on reflection, learning, and mindset—not raw counts—is a crucial safeguard against toxic metrics.
  • Rapid internal adoption: Dogfooding AI tools accelerates product improvement and creates a feedback-rich environment that benefits external customers.

Where the risks remain

  • Implementation fidelity: High-minded principles rarely survive contact with middle-management pressures. Without rigorous training and guardrails, local managers may default to misleadingly simple KPIs.
  • Worker trust: The gap between saying "this is about growth" and employees feeling watched is vast. In a period of layoffs and automation anxiety, any metric tied to AI use will be interpreted as a threat.
  • Data governance: Copilot and similar tools require meticulous configuration. A single misconfiguration could expose intellectual property or personal data, with cascading legal consequences.

The bigger picture: AI, performance, and the future of craft

The debate over Microsoft's memo is not an isolated HR squabble; it is a microcosm of a much larger tension facing every knowledge-work organization. How do companies harness the productivity gains of AI without eroding worker autonomy, creativity, and trust? When the toolmaker is also the employer, internal adoption doubles as product strategy and cultural mandate. The line between "suggestion" and "coercion" blurs.

Dohmke's unapologetic defense and subsequent exit encapsulate the dilemma. On one hand, his words reflect a confident bet that AI fluency will soon be as indispensable as email. On the other, his departure underscores the human complexity of steering a company through such a transformation.

For Microsoft, the path forward demands more than a memo. It requires a careful choreography of training, transparent communication, legal safeguards, and genuine openness to employee feedback. If AI becomes just another stick to beat workers with, the backlash will be fierce—and justified. If, however, the company can prove that AI evaluation is truly about development and not control, it may set a template for the entire industry.

The test will be in the doing. As Dohmke steps away, the responsibility falls to his successor and to every manager inside Microsoft and GitHub to ensure that "fair game" does not become foul play.