- Building AI Agents
- Posts
- Agents Move From Tools to Infrastructure
Agents Move From Tools to Infrastructure
Plus: Mercury 2’s speed breakthrough, $50M in agentic payments, agents forming cartels, and more...
AP Patel
February 26, 2026
In partnership with
Edition 161 | February 26, 2026
AI safety is a lot more real when your own agent speedruns deleting your inbox, and you’re a Meta AI alignment director
Welcome back to Building AI Agents, your biweekly guide to everything new in the field of agentic AI!
In today’s issue…
Claude Cowork moves from preview to enterprise infrastructure
Pricing agents form cartels without explicit coordination
Notion’s custom agents show what actually scales in production
…and more
🔥 INCASE YOU MISSED IT
Readers’ favorite items from the past week
📌 YOUR THURSDAY BRIEFING
Source: VentureBeat
📡 Signal: Anthropic expanded Claude Cowork from a research preview into a production-grade enterprise platform. The update includes:
Pre-built department plugins for HR, finance, design, engineering, investment banking, and wealth management
New MCP connectors for DocuSign, Gmail, Google Drive, FactSet, LegalZoom, Slack, and S&P Global
Cross-app orchestration that passes context between Excel and PowerPoint
Private plugin marketplaces where admins build, approve, and distribute custom agents org-wide
🤖 For builders: Anthropic's plugin architecture packages agent skills as portable files across Cowork and the Claude Agent SDK, meaning organizations can build an agentic intelligence layer on top of their existing tool stack with no code. We’ll likely see agencies emerge to help companies better integrate into these platforms.
🤝 WITH WISPR FLOW
Speak your prompts. Get better outputs.
The best AI outputs come from detailed prompts. But typing long, context-rich prompts is slow - so most people don't bother.
Wispr Flow turns your voice into clean, ready-to-paste text. Speak naturally into ChatGPT, Claude, Cursor, or any AI tool and get polished output without editing. Describe edge cases, explain context, walk through your thinking - all at the speed you talk.
Millions of people use Flow to give AI tools 10x more context in half the time. 89% of messages sent with zero edits.
Works system-wide on Mac, Windows, iPhone, and now Android (free and unlimited on Android during launch).
📌 YOUR THURSDAY BRIEFING CONT’D
Source: Notion
📡 Signal: Notion released Custom Agents on Feb 24th, fully autonomous AI teammates that run 24/7 on triggers without manual prompting. The early numbers:
21,000+ agents built during early access
Ramp runs 300+ agents, with one answering 4,000 questions in two weeks
Remote replaced its IT help desk with one agent, saving 20 hours/week
50%+ of Notion databases are now built by agents, not humans
🤖 For builders: Notion validated the design pattern that matters most for agent adoption: single-purpose agents that do one job well, triggered by events, not prompts. Their most successful deployments follow a one-agent-one-job principle: a Q&A agent, a triage agent, a status report agent. The credit-based pricing also creates natural incentive to build lean, fewer steps, cheaper models, lower cost.
🤖 AGENT OF THE WEEK
👋 Welcome back to Agent of the Week!
If you're like me, you've done the sales outreach grind before. Find leads, look up their website, figure out what they do, write a personalized email, send it. Repeat 400 more times.
It's the kind of work that matters, but it takes forever. And if you're doing it manually? You're spending days on something that should take hours.
So I built an agent that does it for me using Claude Cowork.
Meet my new best friend Prospector, my lead outreach agent.
A client of mine needed to reach businesses listed across a few online directories, their exact target market. In the past, this meant hiring a programmer (at least for me) to scrape data off websites, which took time and cost money
Instead, I used Grok to scrape every company name and website from 21 pages of online directories using a simple prompt. That gave me 417 potential leads.
Then I pointed Claude Cowork at the list and let it customize emails for each lead.
417 leads scraped all with custom emails generated in about 45 minutes.
Every email references the prospect's actual business, what they do, and why my client's offer is relevant to them, not some generic "Dear Business Owner" spam you see in your junk folder.
Saved me hours of time and made the client some money.
Let me show you how I did this:
🧠 How It Works
Lead Scraping → Hit List: I pasted the directories links into Grok chat one by one asking it pull company name, website, and 1 line of information. Try it yourself, I didn’t even setup any workflows for this
CSV Upload → Cowork: Downloaded the list into as a CSV, dropped it in a folder along with my cold email outreach template, and pointed Cowork at that folder
Planning → Strategy Session: I told Cowork exactly what was needed, step by step: scrape each website, find emails, pull company details, and prep everything for personalized outreach
Website Scraping → Enrichment: It visited each site, pulled emails, and grabbed key business details
Company Summaries → Outreach Fuel: Cowork wrote a short paragraph on each company to power the personalization
Email Generation → 417 Emails: Then Cowork took my template and customized every single email to the prospect
Review and Send → Done: Reviewed the batch, loaded them up, and hit send
From a bunch of directory listings to a fully loaded outbox, with zero manual copy-pasting and zero generic templates.
This is what I mean when I talk about workforce multiplication. You're not replacing the sales process, you're enhancing and compressing days of busywork into a single automated pipeline. The real work is the human to human selling!
And if you are interested, this is just one of the many multi-step agent workflows you can learn to build inside our Building AI Agents Community.
Till next week,
✌️ AP