The remote employee experience is changing faster than most HR teams are prepared for — not because the fundamentals of what people need have shifted, but because the tools available to address those needs have expanded dramatically, and the risks of using them poorly have expanded alongside them.
At Running Remote 2026, people leaders from Zapier, Binance.US, and Givebutter shared where they are in that evolution, including what is working, what they built that failed, and what they would do differently. The moderator was Stacey Nordwall, VP of People Strategy at Pyn.
What AI actually does to the employee experience
The opening observation from the panel was precise: AI is raising expectations faster than it is changing processes. First drafts look exponentially better than they did a year ago. The performance bar has shifted. Employees who use AI tools are producing outputs that would have been remarkable a year ago, and teams are adjusting their expectations of what good looks like accordingly.
This creates psychological pressure alongside the productivity gain. People are simultaneously building confidence through AI coaching tools and experiencing anxiety about whether their baseline performance is keeping pace with what AI-augmented work now produces.
One distinctive insight: AI agents developed and customised over months or years could become retention tools. A deeply trained AI workflow that understands a specific employee’s working style, their communication preferences, their area of expertise — that becomes genuinely valuable to walk away from. Companies are beginning to realise that the AI systems employees build are company assets, and the employees who invest heavily in building them may be less likely to leave.
Three implementations worth studying
Binance.US implemented an AI benefits chatbot to create a judgment-free environment for employees to ask questions about deductibles, claims processes, and coverage details. The insight driving it: most employees feel embarrassed asking basic benefits questions they think they should already know, which means they either stay confused or ask a colleague rather than HR. The chatbot removed the social cost of the question. Survey results showed meaningful improvement in benefits confidence and perceived value — without changing any of the actual benefits offered.
Givebutter gave every employee a monthly AI stipend to experiment freely, creating an ‘AI at Work’ Slack channel for sharing what they discover. The energy this produced was notable: their Director of Support vibe-coded a chatbot and got engineering feedback. Cross-functional learning happened organically because a channel existed for it. They also began hiring in eleven concentrated global hubs in fall 2025, deliberately building talent density in specific markets to enable human connection in an increasingly agentic world.
Zapier defined formal AI fluency standards in May 2025 — a minimum performance bar that cascaded into changes across onboarding, performance management, and recognition. They built a digital onboarding buddy using Claude and MCP that creates manager hiring plans, guides new hires through self-paced onboarding, schedules meetings, checks understanding through questions, and provides a safe space to ask questions without social risk. Zapier’s managers run daily AI digests that pull from meeting notes, Slack channels, and team conversations to surface work they would otherwise miss — enabling better championing of team achievements and faster problem identification.
What they got wrong
Three failure modes emerged that are worth naming explicitly.
Over-reliance without review: a manager at Binance.US used AI to write a performance review, which cited incorrect dates and gave a rating based on wrong information without any human verification. The principle broken: you can delegate the work but not the accountability.
Tool sprawl: Givebutter’s original AI stipend produced eight different note-taking apps across a small team. Tool fatigue, budget creep, and the genuine pain of consolidation followed. Offering experimentation without guardrails creates a different kind of chaos.
Automating broken processes: Zapier’s talent team applied automation to their referral system, only to discover the automation accelerated the brokenness rather than fixing it. They stepped back, redesigned the process from its philosophical foundation, then reapplied AI. The lesson: interrogate whether a process is right before automating it. AI amplifies what is already there — good and bad.
Creating space for experimentation
The panel converged on a shared approach to building AI fluency that respects the reality that most people are neither early adopters nor active resisters — they are curious but busy.
Zapier’s talent team runs build days: four per quarter, four-hour protected blocks where recruiters are slightly unavailable to inbound requests. They work within guardrails, pair with colleagues, and show what they built at the end. The accountability is gentle and social rather than evaluative. Team members who were teetering on disengagement have turned around completely.
Givebutter uses lightning talks — five minutes at all-hands meetings for someone to show something they built. The barrier to entry is low. The learning is peer-to-peer, which carries more credibility than top-down training. Lucy Stein’s recommendation was direct: find the person on your team who is already nerding out on Claude, and ask them to explain it to you like you are ten years old. That levelling moment — where a leader shows genuine curiosity rather than authority — changes the culture around AI experimentation faster than any policy.
