Andrey Khusid has a blunt message for executives chasing AI returns. Bolt the technology onto existing teams and processes, and the promised productivity surge will prove elusive. Companies must rebuild workflows from first principles instead. The Miro co-founder and CEO delivered that assessment Wednesday at Wave by Vento in Turin, Italy, barely a month after Bending Spoons agreed to acquire the visual collaboration firm. (The Next Web)

His warning lands at a moment when organizations pour billions into large language models and agentic tools. Individual output climbs. Project timelines often stay stuck. Khusid argues the mismatch stems from a fundamental error. Teams treat AI as an overlay rather than the foundation.

“One of the big challenges is how not to put AI on top of the existing teams or operational systems, but actually, from first principles, design the workflow and then only add humans to add value on top of that workflow,” he said. The quote captures the core of his prescription. Start with the process. Determine where humans add unique judgment. Layer AI everywhere else.

Miro tested the approach internally. In select functions the company created parallel teams that performed identical work using heavy AI assistance. Leaders built new processes around those groups first. Only after validation did they roll the model out more broadly. The experiment revealed gaps that surface only when AI operates inside fresh structures. Context lives scattered across people and documents. Bringing every fragment together becomes the initial task. Without that step, even sophisticated models operate in isolation.

Noise presents another obstacle. Generative tools let quieter voices contribute prototypes quickly. That inclusion boosts idea diversity. Yet the same capability floods inboxes and boards with drafts, summaries, and variations. The result? More content. Less signal. Khusid insists accountability solves the problem. Every individual must stand behind what they introduce. No passing off AI output as finished thought.

Recent data from Miro itself underscores both the opportunity and the coordination challenge. Since February 2026 its MCP server has processed more than 19 million tool calls from AI agents. Monthly calls grew 34.9 times between February and August. User numbers expanded 16.8 times in the same period. September produced roughly 6.3 million calls alone. Fifteen different AI clients now connect, with Claude variants dominating early usage before ChatGPT accelerated. (Miro Blog)

Those agents treat the Miro canvas as both input and output. More than half the calls involve an agent reading board content and acting on it. Another substantial share generates new boards, diagrams, timelines, or prototypes directly on the shared workspace. The pattern suggests a shift from solitary chat windows to persistent, visual collaboration spaces where humans and agents operate side by side.

Jeff Chow, Miro’s chief product and technology officer, highlighted the advantage. When AI produces a deck inside the canvas, teams open it immediately, debate elements, and revise together before any presentation. The output never leaves the context where decisions happen. That integration addresses one of Khusid’s central concerns: individual speed without corresponding team velocity yields limited business impact.

Inside Miro the emphasis has long been experimentation. The company maintains an unlimited budget for learning and development tools. Employees gain essentially unrestricted access to the latest AI offerings. Khusid views the spend not as a line-item expense but as core training. “Our L&D budget is unlimited tooling,” he told Business Insider earlier this year.

Yet he demands a business case for each acquisition. The real metric? Velocity of innovation. Miro tracks projects through a discover-define-deliver sequence and measures how quickly teams advance between stages. Time savings appear across engineering, product, and design. Still Khusid cautions that today’s patterns likely represent an interim state. Real clarity on workplace configuration may not arrive until late 2026 or 2027.

His views align with broader conversations at McKinsey. In a July interview the CEO described a move toward “makers” who span traditional roles. Product managers, designers, and engineers increasingly create, iterate, and ship with AI support. Dependencies remain, but expectations around speed rise. Agentic workflows add another layer. Coordination now includes human-to-agent and agent-to-agent interactions. (McKinsey)

Khusid expects visual decision-making to dominate within three years. Voice and conversation will handle much of the interaction. The canvas becomes the persistent record where reasoning, prototypes, and outcomes live together. Miro’s own product evolution reflects that bet. Features like Sidekicks, Flows, and Connectors pull context from Slack, GitHub, and other systems directly into the workspace. Agents generate content on the board rather than in isolated chats.

But. The technology amplifies existing problems when misalignment exists. As one analyst noted after hearing Khusid speak, AI makes bad processes worse faster. Clarity on vision and strategy must precede heavy automation. Otherwise organizations generate impressive volumes of workslop that demand even more human time to filter.

Accountability therefore sits at the center. Humans remain responsible for outcomes. Agents propose, synthesize, and execute within defined bounds. The canvas makes every contribution visible and editable. That transparency reduces the temptation to treat AI output as authoritative without review.

Startups already operate this way in pockets. Large enterprises experiment with similar models in innovation labs. Scaling the practice across functions requires more than pilot projects. It demands new operating models, updated accountability structures, and metrics focused on end-to-end velocity rather than isolated task completion.

Khusid’s own company offers a case study. Profitable since 2016, Miro could afford to invest aggressively in AI experimentation. That financial cushion allowed leaders to prioritize learning speed over immediate ROI calculations. Most organizations lack the same runway. They must still confront the same question. Does the current workflow deserve AI acceleration, or does the workflow itself need replacement first?

The acquisition by Bending Spoons adds another variable. The deal, announced in September, places Miro under new ownership with deep experience in consumer apps and AI. How that transition shapes product direction remains unclear. Khusid’s public comments suggest continuity on the core thesis. Rebuild work around AI. Keep humans in the loop for judgment and ownership. Connect everything on shared visual surfaces.

Executives listening to the Turin remarks face hard choices. Training programs that teach prompt engineering address only the surface. Real gains require process engineers who can deconstruct existing operations and reconstruct them with agents as primary actors. Job descriptions will change. Some roles may shrink. Others will expand into orchestration of multiple AI systems.

Data from Miro’s MCP server points to rapid adoption among product, design, marketing, operations, and project management teams. Usage grows not just in volume but in sophistication. Agents now produce clickable prototypes, Kanban boards that convert to timelines, and workshop formats complete with dot voting. The canvas turns static AI output into living artifacts.

That evolution matters. When an agent reads the current board state, it operates with team context rather than generic knowledge. When it writes back to the same board, the output enters the collaborative flow immediately. Context switching decreases. Momentum increases.

Still the human element persists. Someone must define success criteria. Someone must resolve conflicts between competing AI suggestions. Someone must accept accountability when the output reaches customers or leadership. Khusid returns to that point repeatedly. Productivity at scale depends on redesign, not augmentation.

His counsel to founders echoes the same discipline. Work on problems that genuinely interest you. Trust your intuition. The AI wave will reward those who question assumptions about how work gets done rather than those who simply deploy the latest model faster than competitors.

Organizations that heed the message may capture the multiple-x returns Khusid describes in McKinsey conversations. Those that treat AI as a productivity overlay risk watching individual metrics improve while enterprise outcomes stagnate. The difference lies not in the technology but in the willingness to redesign from the ground up.