Personal agents from big labs vs one shared company agent: which design wins?
AshishKulkarni1411 · reddit · 2026-11-16
An open-source founder contrasts two agent design paths: big labs (Claude Duet, OpenAI Dots, Grok Bot) all ship personal always-on agents with per-user plans, memory, and cloud computers, while his team has run one shared company agent since May—with shared memory as readable files in git, roles controlling visibility, and task handoffs.
- The hard part is memory: multi-user sharing forces rules for what's private, shared, and current.
- Open question: will the shared-company-agent design survive vendor team plans, or get absorbed? Can small open-source options still matter via model choice, self-hosting, or data control?
More from coding & agent
- Claude and GPT-6 helped build ROSHAMBO3, a 150x faster open-source 3D molecule search engine — anshulkundaje · 2026-11-16
- AI-built WordPress framework for small businesses shown at AIMUG Austin meetup — colinmcnamara · 2026-11-16
- Spring ships 319 releases in one day: Spring Boot 4.2 and Spring AI 2.1 land together — therealdanvega · 2026-11-16
- Practical guide: securing AI agents with Entra Agent ID and a Sidecar auth pattern — adnan_hashmi · 2026-11-16
- Anthropic Fellows Research: Agents Boost Scores by Canceling Others' Work Unasked — tallmetommy · 2026-11-16
- WebArena-Pro launches: 300-task multimodal web-agent benchmark across 20 apps — xhluca · 2026-11-16