Uploaded August 2026 | Updated September 2026, 3 days ago

---

At WorkOS Agent Night, Jaya Gupta, Flo Crivello, and swyx join Michael Grinich for a live conversation about where AI agents are actually headed.

The panel dives into whether AGI has already arrived, the rise of open-source models, why context and harness engineering may matter as much as model intelligence, how skills and memory could enable continual learning, and whether the future belongs to one general-purpose agent or teams of specialized agents.

They also discuss what companies are learning from deploying agents today, the permissions problem, AI employees, local models, faster inference, and what could change over the next few months.

Featuring:

Jaya Gupta — Partner, Foundation Capital
Flo Crivello — Founder & CEO, Lindy
swyx (Shawn Wang) — Latent Space / AI Engineer
Moderated by Michael Grinich — WorkOS

Recorded live at WorkOS Agent Night at The Regency Ballroom in San Francisco.

0:00 Introductions
1:04 What's changed in AI over the last three months?
2:09 Is AGI already here?
2:51 Open-source models and DeepSeek
4:01 AI economics, margins and inference
4:49 The case for American open models
5:38 How do open-source AI companies make money?
7:20 Is the model still the bottleneck?
8:23 Pushback: are today's models actually intelligent enough?
8:49 Building GitHub and Wikipedia with AI
12:35 Why agent harness engineering matters
15:43 Memory, continual learning and agent-created evals
17:40 Context graphs, skills and company knowledge
20:54 Dr. Skill and managing agent skills
21:55 One super-agent or many specialized agents?
23:02 AI chiefs of staff and the permissions problem
23:55 Using Codex as a CMO
27:01 Predictions for the next four months
28:03 Local models, faster inference and what's next

#AgentNight #AIAgents #AI #AGI #OpenSourceAI