New in Claude Code (research preview): dynamic workflows. Claude writes an orchestration script on the fly, then spins up a large fleet of coordinated subagents in parallel to take on your most complex tasks. Use the word "workflow" in a prompt to get started.

Sunday thoughts. A few AI moves on my mind. Cursor wiped a database in 9 seconds. Salesforce: the API is the UI now. Rufus buys without asking. Meta opened ads to AI agents. AWS sold 30 years of ops as agents. Stripe shipped 288 agent products. The pace has shifted.

How does a 150-year-old, 2B vacuum company pivot from 0% Claude to 85% in just a few months? “We were very aggressive in giving employees access to the tools that they needed to experiment.” That’s @matthew_kruer (CIO, Bissell), joining @FoldesCraig (Founder, ChatWalrus) + @BenFlohrOG (Co-Founder, Scale) to share his playbook for training over 1,000 employees on AI. Plus … Agent nukes database 00:02:55 60 Salesforce MCP tools 00:07:56 213 ads scraped in 1hr 00:15:00 Stripe’s 2T bet 00:20:03 700B in AI CapEx 00:24:14 Meet Matt Kruer 00:30:06 0% to 85% adoption 00:33:03 The pit of disillusionment 00:34:31 Building consumer brain 00:37:30 “Will AI take my job?” 00:44:10 AI ads fooled everyone 00:47:53 Base vs. reasoning model 00:52:25

Spent way too long figuring this out. Saving you the math. GPT-5.5 standard: $5 per million input tokens. GPT-5.5 Pro (reasoning): $30 in, $180 out. 6x for the thinking version. Drafts, summaries, tagging: base model. Pricing, code, agent decisions: turn the thinking on.

Stripe shipped 288 products at Sessions last week. Most solve one thing: easy plumbing between merchants and the agents doing the buying. Walmart proved it. Built Sparky, plugged it into ChatGPT, conversion ran ~70% of site. The shelf is inside the agent.

AWS launched Connect Decisions last week. 30 years of Amazon's supply chain playbook, sold as agents. A $50M brand could swap $300K of Blue Yonder plus $150K of consulting for usage-based pricing closer to $100K all-in. Bigger upside: it learns from your planner's overrides.

Missed this last week. Amazon's Rufus now places orders for shoppers — recurring buys, birthdays, price triggers, predicted run-outs. 300M shoppers, engagement up 400% YoY. Rufus picks from product info. Keep titles, bullets, attributes clean or stop showing up.

Two questions I'm asking every SaaS we work with. Can my AI agent do everything your dashboard does? What does the rate limit cost when an agent's running thousands of calls a day? Most of us have never asked either one in a demo. Both are about to be table stakes.

Watching this in real time is wild. Shopify AI Toolkit, Salesforce Headless 360, Klaviyo Custom Skills. Three platforms, three weeks, same playbook: open the backend to agents through MCP and APIs. The API surface is the product now. The dashboard is just a backup.

Sunday thoughts. Five AI shifts shaping operator playbooks. GPT-5.5 retook the leaderboard. Workspace Agents killed Custom GPTs. Walmart fired OpenAI's checkout. Project Deal: agent-to-agent commerce. MCP became the standard plug. Curious which one operators are sleeping on.

Bill Nguyen pays tens of thousands a month to run Claude, ChatGPT, and Gemini on his own data. Says he's paying for education. You don't need $30K. Buy a GPU with the most VRAM you can afford. Run an open source model on it. A weekend of that beats a year of newsletters.

Meta cutting 14,000 roles. AI capex up from $72B to $135B. Microsoft, Amazon, Block, Oracle running the same playbook. Big tech was overstaffed before AI showed up. Smaller brands usually aren't. For the rest of us, the move is being more thoughtful with the next hire.

Every brand is using AI to write copy, create ads, and close tickets. But where is the alpha? @mbertulli (CEO, Pela Case + Lomi) and Arias Giri (Head of Ops, BlockBlueLight) join hosts @FoldesCraig (Founder, ChatWalrus) and @BenFlohrOG (Co-Founder, Scale Media) to discuss how AI is changing ecom … and reveal their secrets. - 16k SKUs, made on-demand - 6,000 hours saved per year - 69 agents closing 190 deals - 35% AOV lift with Sparky - 92k tech jobs gone (14k Meta) “AI is not coming for your job, but it’s coming for the people that aren’t using it.” Full episode ↓ Claude vs. GPT 5.5 00:03:11 Walmart fires OpenAI 00:11:39 Meta cuts 14K jobs 00:20:26 How to clone yourself 00:24:50 69 agents, 190 deals 00:28:57 16k on-demand SKUs 00:34:46 1 person, 8 brands 00:39:46 Stop thinking alone 00:45:12 WTF MCP in 5 minutes 01:09:35

MCP is the standard connector between AI and your business data. Each SaaS builds the integration once. Any AI plugs in. When something changes, they fix it. Ask every SaaS in your stack: do you have an MCP server. The yes/no answers tell you who's still around in 18 months.

Anthropic gave 69 employees $100 each. Claude agents traded personal items in Slack. 186 deals. Half Opus, half Haiku. Same broken bike: Opus $65, Haiku $38. Whichever side runs the better model wins. The losing side never notices. Which model are you using most?

This is still the best f****** resource on the internet about landing page fermatcommerce.com/ecommerce…

The cost line on GPT-5.5 is being misread. Half the cost of Opus 4.7 head to head, sure. If you're already on GPT-5.4 and upgrade, ~20% more per task. API doubled, tokens dropped. Two different conversations.

Custom GPTs are being deprecated. Workspace Agents replaces them on Business and Enterprise. Free until May 6, then credit-based pricing OpenAI hasn't disclosed. Build-once-share-with-the-org is the next era of internal AI tools. Pilot before pricing lands.

Weekend thought for operators. Phase 1: use AI to do existing work faster. Phase 2: use AI to cut cost. Phase 3: do something your company literally couldn't do before AI existed. PwC says 20% of companies capture 74% of AI value. Guess which phase they're in.

.@codyplof rebuilt Jones Road’s entire cart last week — from his phone, in the Bahamas. @Bbruhis built an AI agent to find every Costco buyer’s email + cold pitch them to bust into retail. How? That’s what the newest episode of AI Ops, with @FoldesCraig and @BenFlohrOG, is all about. From the operators on the bleeding edge … - @krishnapoda’s AI data-house - Bar’s 1-star review playbook - Claude Design for landers - Klaviyo AI for entire campaigns - Ben’s three-phase AI rollout Full episode ↓ OpenAI Ad Manager 00:03:30 Anthropic’s Billing Shift 00:06:59 Klaviyo AI Campaigns 00:11:22 PwC: 20% Capture 74% 00:13:36 Claude Design vs Figma 00:16:49 Claude Code Workflow 00:25:51 Why Roles Are Converging 00:33:41 CEOs Must Lead AI 00:43:03 1-Star Review Playbook 01:10:13