Two Thresholds Crossed

In partnership with

Hey Friends,

Two big shifts landed this week. Kling’s Motion Control lets anyone paste realistic motion onto any face; deepfakes just got easy. Claude’s Cowork turns AI from chat assistant into an actual coworker with file and browser access. Both are live now.

Let’s get into it.

Help us make better ads

Did you recently see an ad for beehiiv in a newsletter? We’re running a short brand lift survey to understand what’s actually breaking through (and what’s not).

It takes about 20 seconds, the questions are super easy, and your feedback directly helps us improve how we show up in the newsletters you read and love.

If you’ve got a few moments, we’d really appreciate your insight.

Take the survey.

Kling Motion Control: The Deepfake Threshold

Upload a photo, pick a motion template (dance, expressions, gestures), and generate. Kling blends your face onto reference motion—realistic output in seconds.

The breakthrough: Pre-built motion sequences instead of wrestling with prompts. Pick “Expression Challenge” or “Ghost Step Dance,” match orientation, and go. Motion transfer with guardrails.

The quality jump—stable hands, consistent expressions, multi-second shots—crosses from “obvious AI” to “plausibly real.” That’s a different category of risk.

Practical rules if you use it:

• Reference quality > prompt detail. Use clean, well-lit videos—flaws transfer.

• Match orientation early. Left-facing subject + right-facing motion = artifacts.

• Prompt for vibe, not motion. Motion comes from video; text prompts handle lighting and mood.

⚠️ The reality check: This is a deepfake tool. Celebrity impersonation videos are already circulating. The EU AI Act classifies deepfakes as high-risk; U.S. states are adding criminal penalties. If you build with this, assume consent and verification will become legal requirements.

Claude Cowork: AI as Actual Coworker

Cowork gives Claude access to a folder on your computer. It can read, edit, create files—and with Chrome, navigate browsers too. Available in macOS for Claude Max subscribers.

Think of it as managing a small AI team—multiple parallel agents with defined roles. I run 2–3 Cowork sessions alongside 2–3 coding agents, all on Opus 4.5.

How I actually use it super helpfully like this; Best one at the end:

• Organizing messy folders
  "Analyze all files in this folder. Create subfolders by category (Images, Documents, PDFs, Screenshots, etc.). Move and rename files logically (add dates to screenshots, descriptive names). Remove duplicates and summarize changes."

• Creating expense reports from receipts
  "Scan all receipt images and screenshots in this folder. Extract date, merchant, amount, and category for each. Create an Excel file 'Expenses.xlsx' with columns: Date, Merchant, Amount, Category, and row/column totals."

• Building presentations from notes/assets
  "Review all notes, text files, and images in this folder. Extract key points, create a structured outline, and generate a branded .pptx file with titled slides, bullet points, and inserted visuals."

• Generating reports from scattered notes
  "Read all notes and documents in this folder. Identify main themes, summarize insights, and create a structured report (Markdown or Word) with sections: Executive Summary, Key Findings, Recommendations."

• Financial statement analysis (e.g., spotting inefficiencies)
  "Analyze all financial statements, bank exports, and transaction records in this folder for December 2025. Identify spending patterns, duplicate or unnecessary expenses (e.g., multiple subscriptions/licenses for the same service), overspending categories, and financial inefficiencies. List critical tasks needing attention (e.g., cancel redundant Microsoft licenses), provide actionable recommendations, and summarize overall financial health." → follow up with cancel subscription xy, etc.

Browser reliability: ~80–90% in practice. Navigation and data extraction work well. Complex form submissions occasionally need a retry. I treat it as “draft mode”—I review outputs before acting on them.

Model note: Opus 4.5 is best for this—80.9% on SWE-bench, strong multi-step reasoning. Tradeoff: $5/$25 per mil

The paradigm shift: Not “ask AI a question” but “assign AI a task.” Define workspace, set constraints, let it operate. That’s orchestration, not prompting.

Quick Hits

• OpenAI’s Codex reportedly adding similar folder-access capabilities in testing. The pattern is becoming table stakes.

• Microsoft announced Copilot Agents for M365—browser + file access, action-taking. Everyone’s racing to the same destination.

One thread connects these: agents are becoming workflows. Not “ask AI a question” but “assign AI a task and verify results.” The skill isn’t prompting anymore—it’s orchestration.

That's all for this week!

Happy Building!
🙇Martin

• I recommend:

  • Beehiiv if you write newsletters.

  • Superhuman if you write a lot of emails.

  • Cursor if you code a lot.

  • My book - Generative AI: Navigating the Course to AGI.

  • Follow me on X.com.

• AI for your org: We build custom AI solutions for half the market price and time (building with AI Agents). Contact us to know more.