For the first time, anyone running Facebook or Instagram ads can connect Claude directly to their ad account and manage campaigns through plain English. Read live performance data. Identify failing creatives. Generate new ads. Launch new campaigns. All without leaving the conversation.

A fitness brand running this workflow reported a 15% monthly revenue lift in the first month after deployment, traced to a single morning routine where Claude identifies creatives with declining CTR, suggests new copy and headline combinations, and creates the ad sets directly.

Most marketers pay a media buyer $1,500-$3,000/month for the mechanical parts of this work. Most small business owners do it themselves at the cost of 5-8 hours a week. Both situations are about to change.

This week: the complete 3-prompt loop that runs your ad operation from one Claude conversation, plus a $902 surprise hitting GitHub Copilot users on June 1 you need to know about.

The Old Way:

You run Facebook or Instagram ads. Either you do it yourself, or you pay someone $1,500-$3,000/month to do it for you. Either way, the work looks the same: log into Ads Manager every morning, download a CSV, stare at columns of CTR and ROAS, try to figure out what's working, brief a designer for new creatives, wait 3-5 days for the files, upload them manually, set the targeting, hit publish, wait another week to see if the change worked.

The mechanical parts of this job (pulling the data, identifying winners, briefing creatives, launching new campaigns) consume 70% of the time. The strategic parts (which audiences to test, which positioning to use, when to scale) get whatever time is left over.

The best media buyers are worth their fee. They make creative decisions you can't replicate. But most small businesses aren't paying for senior creative strategy. They're paying $1,500-$3,000/month for someone to do the mechanical work because they don't have time to do it themselves.

The Replacement:

3 workflows. One Claude conversation. The complete loop that replaces the mechanical work.

On April 29, 2026, Meta launched its official Ads MCP server in open beta. The MCP gives Claude direct read-and-write access to your ad account through 29 tools covering the full Marketing API: campaigns, ad sets, audiences, creatives, catalogs, and diagnostics. According to Meta's official launch, the connectors are built and maintained by Meta and use their own authentication system, eliminating the account suspension risk that came with third-party connectors.

The loop runs like this:

Step 1 (Morning Audit): Claude reads your live campaign data, flags creatives with declining CTR or rising frequency, surfaces top performers, and identifies the patterns that separate winners from losers.

Step 2 (Creative Brief): Claude analyzes the winning ads and generates a brief for new creatives matched to the same pattern. Hook structure, visual style, copy formula, target audience.

Step 3 (Launch): Claude writes the new ad copy, creates the ad set in your Meta account in paused state, and queues it for your approval before going live.

Total time per morning: 15 minutes. The system runs every weekday. You approve. Meta launches the campaigns.

What this replaces: a $1,500-$3,000/month media buyer doing the mechanical work, or 5-8 hours of your own time per week if you've been doing this yourself.

Setup time: 15 minutes, no coding required.

The Pro section below includes:

What you get out of this: a complete ad operation running in 15 minutes per morning. Saves 5-8 hours per week if you've been doing this yourself, or replaces $1,500-$3,000/month if you've been paying someone.

"I set up one automation from the toolkit last weekend and saved 6 hours in the first week alone. It feels like having an extra team member running in the background."

Elena M. DTC Founder.

🔒 If you want the complete 3-prompt loop running on your Meta account by Friday afternoon, the full workflow is below. Setup is 15 minutes. The morning routine is 15 minutes per day. Pays for your Claude Pro subscription thousands of times over.

Google announced Gemini Spark at I/O on May 19. It's a 24/7 cloud-based personal agent that runs continuously in the background, even when your devices are off. The agent operates on Gemini 3.5 Flash with Google Antigravity 2.0 orchestration, and runs on dedicated Google Cloud virtual machines rather than your local hardware.

Why this week: Most AI assistants stop running when you close the tab or shut your laptop. Spark doesn't. It continues processing tasks, monitoring your connected accounts, and producing output while you sleep. Tomorrow morning, the work is waiting for you.

Spark integrates natively with Gmail, Google Docs, Slides, and the rest of Workspace. It supports MCP for external connections to Canva, OpenTable, Instacart, and other tools. The agent can handle multi-step tasks that previously required manual orchestration: draft three pitch decks based on your notes, book a restaurant for Friday with these constraints, summarize this week's competitor coverage.

🔒 Implementation checklist is for pro members. Get a breakdown plan to implement AI into your workflow.

🚨 GitHub Copilot goes metered June 1. GitHub announced all Copilot plans will transition to usage-based billing on June 1, 2026, replacing Premium Request Units with GitHub AI Credits at $0.01 per credit. Code completions remain unlimited. Everything else is metered. (GitHub Blog)

🌟 Google launches Gemini 3.5, Spark, and Pics at I/O. Sundar Pichai's keynote on May 19 opened with the milestone that Google's APIs now process 19 billion tokens per minute and 375 Cloud customers processed more than one trillion tokens each in the last 12 months. The headline launches: Gemini 3.5 Flash as the default Search model, Spark as the 24/7 personal agent, and Pics as the AI-powered design app inside Workspace. (Google I/O 2026)

📣 OpenAI launches self-serve ChatGPT ad platform. OpenAI opened a self-serve ad platform that lets advertisers place sponsored responses inside ChatGPT conversations. Two implications: ChatGPT outputs may now be influenced by paid placements, which affects how you should verify recommendations. And if you run any business that depends on online discovery, this is a new acquisition channel to evaluate. (Tech Insight)

🐉 Google DeepMind releases Gemma 4. A family of four open models (E2B, E4B, 26B MoE, 31B Dense) purpose-built for advanced reasoning and agentic workflows, all licensed under Apache 2.0. The 31B model ranks as the #3 open model globally on the Arena AI text leaderboard. Why this matters for you: For high-volume work where token costs matter (the Copilot story shows where this is going), open models running locally or on cheap infrastructure are no longer a fallback. (Dentro AI News)

GitHub announced on April 28 that all Copilot plans will transition from request-based billing to usage-based billing on June 1, 2026. Premium Request Units, the fixed-cost abstraction that let you run unlimited agent sessions for the same price as a quick chat, are gone. In their place: GitHub AI Credits, where every token you send to a model is metered and billed at published API rates. One AI Credit equals $0.01.

The monthly plan prices stay the same. Copilot Pro is still $10/month, Pro+ is still $39/month. What those plans actually buy you is a different question.

One developer reported on the GitHub community forum that their April 2026 usage, estimated at $39.07 under the old PRU model, came back at $902.72 in the preview billing tool. Not a typo. The old model charged the same for a 30-second code fix and a 2-hour autonomous refactor. The new model does not.

GitHub's chief product officer Mario Rodriguez explained the rationale in the announcement: "Today, a quick chat question and a multi-hour autonomous coding session can cost the user the same amount. GitHub has absorbed much of the escalating inference cost behind that usage, but the current premium request model is no longer sustainable."

The translation: developers using Copilot mostly for autocomplete will see almost no change. Developers using Copilot for agentic coding, code review, or complex reasoning are about to see bills they didn't expect.

Six days until June 1. Most Copilot users haven't checked their preview bill yet.

Free across all plans (unlimited): Code completions and Next Edit Suggestions. The original inline-suggestion experience. If you only use Copilot as you type, your bill looks identical to today's.

Metered (consumes credits): Chat, agentic coding sessions, Copilot CLI, Copilot Spaces, Spark, code review, and any third-party coding agents using Copilot.

Double-billed: Copilot code review is metered for AI Credits (the model tokens) AND for GitHub Actions minutes (the infrastructure that runs the review). Self-hosted runners avoid the Actions charge.

The model multipliers vary widely. A quick question to a lightweight model might cost 5-10 credits. A long agentic refactor across multiple files using a heavy reasoning model can easily run 1,000+ credits in a single session.

🔒 The full breakdown (the 3 specific behaviours burning credits fastest, how to cut your bill 40-70% without losing capability, and the alternative stack for Copilot defectors) is for paid members. Start your free 7-day trial.

New This Week:

Prompt Library Updates:

Resources This Month:

Each resource lives permanently in your Pro account. Use them whenever you need them.

Anthropic just gave every small business owner a free AI employee. Most owners haven't heard about it yet. The ones who install it by Monday morning will operate at a structural cost advantage their competitors don't have.

If you do one thing before next Thursday, install the plugin and run /cash-flow-snapshot. The 15 minutes is worth more than the next book you'll read.

👇 What’s behind the paywall:

Pro members deploy AI in their work an average of 5-8x more often than free readers (based on reply data from past issues). The difference is having the exact setup, not the concept.