Perplexity Reveals Agent Strategy

Edition 165 | March 12, 2026

Nothing like a new agent benchmark to get the blood pumping!

Welcome back to Building AI Agents, your biweekly guide to everything new in the field of agentic AI!

In today’s issue…

• Perplexity launches full agent infrastructure stack

• Amazon blocks Perplexity shopping agent in court

• Meta acquires Moltbook, the social network for AI agents

• Nvidia releases Nemotron Super 49B for agents

• a16z argues context engineering is agents’ biggest challenge

…and more

🔥 INCASE YOU MISSED IT

Readers’ favorite items from the past week

1. Anthropic releases a new labor market research report

2. OpenAI launches GPT-5.4

3. Google makes Gmail, Drive, and Docs agent-ready

4. Databricks launches a RAG agent for enterprise search

5. A Google PM open-sources an Always On Memory Agent

📌 THE THURSDAY BRIEFING

Perplexity goes all-in on agent infrastructure at its first developer conference

📡 Signal: At its inaugural Ask 2026 developer conference today, Perplexity announced three things at once. First, Perplexity Computer, their cloud-based multi-model agent that launched two weeks ago, is now available to enterprise customers with SSO, compliance controls, and native Slack integration. Second, Personal Computer: software that turns a spare Mac mini into a locally controlled agent with full file and app access, an audit trail, and user confirmation required for every action, a direct shot at OpenClaw. Third, a rebuilt API platform as a model-agnostic, four-API suite for developers:

• Agent API: a managed runtime for multi-step orchestration across frontier models, with configurable tool access, step limits, and token budgets

• Search API: direct access to the same 200B+ URL index powering Perplexity, refreshed in real time and currently SOTA on SimpleQA and SEAL benchmarks

• Sandbox API: deterministic code execution for agents, coming soon as a native tool inside Agent API

• Embeddings API: anchored by pplx-embed-v1-4B, which leads MTEB retrieval and ConTEB against 30M+ documents

🤖 For builders: This is one of the smartest strategic moves Perplexity could make. As Claude, GPT, and Gemini got good enough to answer more search-like questions on their own, the obvious question became: what exactly is Perplexity's durable advantage? This is their answer. In a single day they shipped a cloud agent, a local device agent, an enterprise agent, and the developer infrastructure to build on top of all of it, without owning a single frontier model. Perplexity is betting its real advantage is not the chat interface, but the infrastructure behind it: live web index, retrieval, orchestration, and eventually execution. If that bet works, Perplexity stops looking like just an AI search engine and starts looking like a core layer in the agent stack.

🤝 WITH HEALTHCARE BREW

Healthcare news for decision-makers

Knowing the healthcare headlines is easy.

Understanding what they mean for the business? That’s the hard part.

Healthcare Brew is a free newsletter breaking down the forces shaping the healthcare industry—from pharmaceutical developments and health startups to policy shifts, regulation, and tech changing how hospitals and providers operate.

No clinical deep dives. No overstuffed jargon. No guessing what actually matters. Just clear, focused coverage built for the people making decisions behind the scenes.

Join 135K+ administrators and healthcare professionals staying informed, for free.

Check it out

 🤖 AGENT OF THE WEEK

👋 Welcome back to Agent of the Week!

If you use a meeting recorder like Fathom or Fireflies, your notes are technically all there. But "technically all there" isn't the same as useful. You're either rewatching the recording or copy-pasting transcripts into ChatGPT just to figure out what you agreed to.

And if you have multiple meetings that day, it gets worse. These tools don’t look across all your meetings from that day and tell you what you actually need to do next. You're still bouncing between tabs, piecing together a mental to-do list from four different call summaries.

And by 5pm your brain is fried. The last thing you want to do is go through this again anyway.

So I built an agent that does it for me in just a few minutes.

Meet the End of Day Meeting Summarizer.

Here's the setup: I have a normal workday. Two to four calls, a mix of client check-ins, strategy sessions, and internal meetings. Each meeting gets recorded. Each one has action items buried in it somewhere.

Normally I'd spend the last 30 minutes of my day piecing it all together: what did I promise that client? What did my team say they needed from me?

Instead, at 5pm sharp, a single email hits my inbox with everything organized and prioritized. And something I can reference the next morning.

🧠 How It Works

1. Meetings Happen → Recording Saves to Your Desktop: Every call gets recorded and the transcript or summary file saves to a folder on your desktop. Nothing new here

2. Point Cowork to the Folder → One-Time Setup: In Claude Cowork, I point it to the folder where your meeting transcripts are saved

3. Open Cowork → Run It Whenever You Want: Whenever I’m ready (mid-day, between calls, or end of day) I just click a button to run it. Cowork pulls every transcript in the folder and reads through all of them

4. Claude Compiles the Plan → Three Buckets: Everything gets summarized and organized into 🔴 Do Tonight, 🟡 Due Tomorrow, and 🟢 This Week, prioritized across all my meetings, not just one call

5. Review → Sent to Gmail: Claude compiles the full summaries and follow-ups and sends it straight to my Gmail to review whenever I want

From four separate meeting summaries to one summarized action plan, delivered automatically at the end of every day.

This can be set up in a tool like n8n, but when I want a quick agentic automation Claude Cowork makes it easy for me to set this up in 5 minutes with just zero code.

And if you want to learn how to build AI workflows like this yourself, come join us inside the Building AI Agents Community.

Till next week,
✌️ AP