I ❤️ smol open models and today @Get_Writer launched the palmyra-mini family with thinking variants! 52.6% on Big Bench Hard (palmyra-mini) 82.9% on GSM8K (thinking-a) 92.5% on AMC23 plus top scores across AIME24, GPQA, and MATH500 (thinking-b). Links below to try them out!

The future firm won’t just execute. It will think. Learn. Evolve. And the enterprises that embrace this shift will lead, compounding intelligence with every cycle. Read the full essay:

As AI‑native employees rise: - 90% of work is handled invisibly by agents - Humans see briefs, recommendations, and decision points - Work feels like flow, not friction

🔹 In the old world: - Employees hunted for tasks across apps and meetings - Coordination lived in Slack and inboxes 🔹 In the new world: - The graph is the interface - Work finds employees - Humans engage only for judgment or creativity

Here’s how it works, like an AI Operating System for the enterprise: - Captures business intent (from humans, systems, and markets) - Scopes and routes work across humans, agents, and tools - Learns and improves with every cycle

AI brings the orchestration graph to life: - It can observe itself - It can route and scope work dynamically - It can learn and adapt The firm becomes a living brain - thinking, orchestrating, and improving continuously.

I call this hidden structure the orchestration graph: - It turns intent into impact - It connects humans, agents, and tools - It’s why big companies function—and why they slow down For decades, it was static and fragile, living in org charts, emails, and meetings.

Every enterprise runs on an invisible network of handoffs, approvals, and escalations. It’s the engine of coordination, and the bottleneck of execution. AI is about to change that. Introducing The Living Brain of the Enterprise 🧵👇 https://t.co/e9vTPdLOi7

7. I wrote about how this changes: - firm structure - performance measurement - strategic advantage 📘 Read the essay →

6. The core question for modern firms isn’t “How many people do we need?” It’s: “What work must be supervised, and who or what should do it?”

5. This isn’t a future scenario. It’s already happening: - PMs coordinating fleets of agents - Ops teams designing control flows, not SOPs - AI-native orgs shipping faster with fewer people

4. Org charts don’t capture that. What does? The orchestration graph: a dynamic network of humans + agents connected by delegation and escalation logic.

3. So the firm is no longer a labor aggregator. It’s a system for orchestration, supervision, and adaptation.

2. But now, one of the core constraints—execution capacity—is vanishing. Agents don’t sleep. Marginal cost is near-zero. Scale becomes software.

1. For 100 years, we built companies around their constraints: labor, coordination, communication.

Execution is free. Supervision is expensive. AI didn’t just change how work gets done—it changed what a company is. Introducing: The Orchestration Graph → a new theory of the firm 🧵

You’re not just doing your job. You’re supervising AI. Delegating. Tuning. Managing. Welcome to the hybrid workforce - where everyone’s a manager now. My new post breaks down: • Human managers • Agent managers • Hybrid managers • Why orchestration is the new core skill

If you’re building with agents, or planning to… This is for you. Read here → https://t.co/0DDjSeDaTw DMs open if you’re working on this problem too.

In the post, I cover: - Why traditional monitoring fails - How to design supervision UX - The rise of the “AI Supervisor” as a new org role What we’re building at @Get_Writer

Agents don’t just return data. They act. They access systems. They ping customers. When things go wrong, a log isn’t enough. You need to know: → Was this behavior safe? → Should it have happened at all?