This insight summarises Is OpenAI's Dots Making ERP and SaaS Obsolete?, by Wei Manfredi, Senior Vice President of AI and Architecture at IHG Hotels & Resorts and an AIHA advisor, published on Substack on 3 October 2026 as part of The AI Disruptor Playbook series.
From interface to action. On 29 September 2026, OpenAI launched dots: always-on agents with their own cloud computers, connections to more than 4,000 apps and, in the enterprise beta, their own corporate identity and credentials. Manfredi had already argued that intent is becoming the front end. Dots go further: agents take over the sequencing, execution and routine judgment that people used to supply between applications, and escalate only the exceptions.
One stay, no human in the loop. A traveler asks a personal agent such as Meta's Muse or Instinct for a Chicago hotel near a client's office. It books and pays on the hotel's own website. A front-desk agent then picks up the reservation, reads the guest's profile (high floor, air conditioning off, slippers), locks a room in the PMS, adjusts the room through building controls and updates the housekeeping and room service lists. Staff see one line: room ready, preferences set.
What survives and what is exposed. Systems of record (reservations, contracts, spend, audit trails) remain and may be worth more, because agents need authoritative data to act on. Workflow engines, approval screens, task queues, RPA and portals built for a person's next step are most exposed.
Whoever owns the action owns the choice. Seat-based pricing stops measuring value when agents do most routine work, and the agent runtime becomes the control plane across vendors. Governance must cover actions, not just access: who is accountable when an agent approves a payment, and what happens when valid permissions still create a confidentiality problem?
What to do now:
- Give agents governed APIs and MCP servers instead of letting them scrape screens.
- Make every agent action traceable: which agent, which system, on whose behalf.
- Write action rights as policy, then encode them where agents run, including identity, loyalty recognition and rate rules for guests' agents.
- Measure how much work and how many transactions agents complete correctly, not how many staff use AI tools.
- Add an agent clause to SaaS contracts before renewal.
- Run one platform that shows all agent traffic on your systems.