monday.com Rebuilt Its Platform Around Agents After Diagnosing an 'AI Dust' Ceiling

monday.com rebuilt its work-management platform around Claude-powered agents embedded directly in workflows after its leadership diagnosed what they internally called an “AI dust” ceiling — surface-level AI additions that never moved usage or outcomes. Since launching the redesigned, agent-first platform in May 2026, the company reports handling more than 5 million agent interactions across IT ticket triage, HR recruitment, marketing competitive intelligence, and executive operations. CPTO Daniel Lereya put the shift plainly: “It meant fundamentally reimagining what the platform should do, not just adding AI to existing workflows.”

The harder lesson in the case study is what scaling actually required: parallel investment in data systems, governance, and permissions, not just the agent layer itself. That’s the step most agent rollouts skip, and it’s usually where a pilot stalls before it reaches production.

Slack ran into the same wall from a different angle. Chief Product Officer Jaime DeLanghe’s framing is blunt: “conversation doesn’t turn into knowledge” without deliberate systems built to capture it. Her fix leans structural rather than technical — defaulting channels to public so agents and humans share context, scoping agents to specific roles instead of one general assistant, and managing human-to-agent handoffs on purpose instead of letting them happen ad hoc.

A more contrarian data point on what “working” looks like at scale: Owner, the vertical SaaS platform for restaurants, says more than 83% of its new customers now start inside its free AI product, up from zero two years ago, and co-founder Adam Guild argues the real signal of a mature AI product isn’t rising logins — it’s falling ones, because well-built software shouldn’t need a human to keep checking in on it.

Three companies, three different functions, one shared thread: the agent layer is the easy part to ship. The governance, data, and role design underneath it is the part that determines whether “agent-first” is a real architecture or just AI dust with better branding — and it’s usually where a consulting engagement finds the actual work.