Investor: Everyone Will Use Two AI Assistants, and Meta and Microsoft Are Already Positioned
RihardJarc · x · 2026-09-28
Investor Rihard Jarc argues the AI assistant market is not zero-sum: each person will eventually want at least two assistants, one for personal life and one for work.
- Personal assistant: optimized around your life goals and preferences, high on privacy, not meant to be observed, audited or transferred. The best IQ model matters less than human-assistant skills — knowing the boss's preferences, holding the calendar and inbox, knowing when to interrupt, being fast, discreet and reliably available. Meta is well positioned thanks to its social algorithms' existing grasp of user preferences.
- Work agent: owned and shared with your employer, deprovisioned when you leave, with memory staying with the company — so no personal data should live there. Employers want it observable and auditable at all times and will reject many connectors a personal agent would want, due to data and cyber risk.
- Interaction: the two agents communicate only where needed (e.g. calendar scheduling) while keeping their data siloed.
Jarc expects companies to avoid depending on a single model vendor, making an orchestration layer between models and model providers almost necessary. Hyperscalers, especially Microsoft, are best positioned: they have the human-tool integrations assistants need plus established enterprise trust, and don't need to own a frontier model — just be the best router and keeper of context and data (the harness). He also shared an image illustrating Meta's Muse potentially talking to Microsoft's Autopilot.
More from AGI Musings
- Polymarket Puts 18% Odds on Frontier AI Labs Agreeing to a Joint Pacing Agreement in 2026 — Polymarket · 2026-09-28
- Survey: 36.3% of global public already believes AI understands or seems conscious — dioscuri · 2026-09-28
- 38% of US counties lack a general surgeon; AI seen as key to expanding clinician capacity — HealthcareAIGuy · 2026-09-28
- If parents face charges for neglect, why don't AI companies for cyber failures? — kevinnbass · 2026-09-28
- Commenter points out a strange double standard: sanction nuclear states, but let risky AI labs shape laws — kevinnbass · 2026-09-28
- Open Models Now 4 Months Behind Frontier, Handle 56% of Vercel Gateway Tokens — bigdata · 2026-09-28