This week was an exciting one. The world came to a halt, and everyone started talking about a single app: Clawdbot, or Molt, or whatever it’s called these days. It was a monumental announcement for an incredible bot that offers a wide range of features. I’ve installed it on a Cloudways VPS to kick the tires. There are countless stories about it, so I won’t add to the already extensive coverage in this newsletter.

In my opinion, this is potentially the most significant breakthrough in functional AI we’ve had so far. This is a peek at what autonomous AI might truly feel like. However, the significant security risks associated with it are a major concern. I’ll write a more detailed post later in the week, providing some tips on how and how not to use it (assuming we’re still alive and the bots haven’t conspired to eliminate us all on their social networks).

Let’s get on with the show…

⚡ Quick Hits

What 100K+ Engineers Read to Stay Ahead

Your GitHub stars won't save you if you're behind on tech trends.

That's why over 100K engineers read The Code to spot what's coming next.

  • Get curated tech news, tools, and insights twice a week

  • Learn about emerging trends you can leverage at work in just 10 mins

  • Become the engineer who always knows what's next

!-- MODULE B: TOP AI UPDATES THROUGH SIGN-OFF -->

🚀 Top AI Updates

Google brings agentic "Auto Browse" into Chrome via Gemini

Gemini in Chrome can now complete multi-step web tasks like research, form filling, and scheduling with its new "Auto Browse" feature. Google also positioned Universal Context Protocol (UCP) as infrastructure to let agents connect across services.

  • Available in Chrome Canary and Dev channels for Gemini Advanced subscribers
  • Auto Browse handles multi-tab workflows, form completion, and context switching
  • UCP aims to standardize how agents share context between tools and services

Why it matters: The browser is the new operating system for agents. Whoever owns the "click layer" wins distribution for real work automation.

Microsoft ships Agent Mode in Excel (Copilot)

Agent Mode transforms natural language outcomes into workbook edits, including formulas, charts, and structural changes. This reduces the gap between analysis intent and spreadsheet execution for non-Excel power users.

  • Rolling out to Microsoft 365 Copilot subscribers in Excel for web and desktop
  • Handles complex tasks like pivot tables, conditional formatting, and advanced formulas
  • Understands business context and suggests data transformations automatically

Why it matters: Spreadsheets remain the lingua franca of business. Agentic Excel is one of the fastest paths from "AI demo" to "AI saved me 3 hours today."

Anthropic adds interactive "Tools" inside Claude (MCP Apps)

Claude now embeds interactive app surfaces like tasks and design files directly in chat, enabling action instead of just conversation. This leverages MCP as connective tissue for real work, not just chat completions.

  • Available in Claude desktop and web apps for Pro and Team subscribers
  • MCP connectors enable direct integration with Slack, Figma, Linear, and dozens more
  • Interactive widgets let you complete tasks without leaving the chat interface

Why it matters: "Agent" stops meaning "writes a plan" and starts meaning "moves the work forward inside the tools you already use."

Perplexity rolls out Learn Mode broadly + Comet upgrades

Learn Mode expands to more users with guided, stepwise explanations. Comet, their agentic browsing experience, received backend model upgrades and enhanced capabilities.

  • Learn Mode now available to all Pro subscribers with expanded topic coverage
  • Comet upgraded to newer reasoning models for complex multi-step tasks
  • Enhanced citation accuracy and source verification in both modes

Why it matters: Search is being re-bundled into "tutor + agent." That's a direct threat to how knowledge work starts and who gets paid for it.

🛠 Pro Tip: "Folder-to-deliverable" marketing ops with Claude Cowork

Use Cowork to transform a messy folder of briefs, notes, and screenshots into a finished weekly deliverable with campaign recap and next steps. This creates repeatable structure that eliminates manual synthesis work and ensures consistent executive reporting.

Example Prompt:

You have access to /Weekly-Marketing/2026-01-W4/. Create:

(1) a one-page exec summary
(2) a KPI table
(3) 5 learnings
(4) 5 next actions
(5) risks

Use the same headings every week. Ask only if something is truly missing.

Why it matters:

  • Eliminates 2-3 hours of weekly manual synthesis and formatting work
  • Creates institutional memory by standardizing how campaign data is captured and analyzed

💡 Productivity Gem: Slackbot as your personal agent

Turn your Slackbot into a personal AI assistant that drafts messages, schedules reminders, and summarizes threads while respecting your workspace permissions. This keeps AI assistance inside your existing workflow without context switching.

Setup steps:

  • Open Slack and search for "Slackbot" in direct messages
  • Enable AI features in your workspace settings (Admin may need to approve)
  • Start conversations with prompts like "Draft a reply to the last message in #marketing"
  • Use it to summarize threads, create follow-up reminders, or draft announcements

Why it matters: AI assistance without leaving Slack means faster execution and less tool fatigue while maintaining security boundaries.

⚕ AI-Enabled Health Tip: Build a medical timeline from your records

Use Claude Cowork to transform a folder of medical PDFs (lab results, visit summaries, discharge notes) into a comprehensive dated timeline with anomalies flagged and follow-up questions prepared for your next appointment.

  • Upload all medical documents to a dedicated folder on your computer
  • Ask Cowork to create a chronological table with dates, tests, results, and provider notes
  • Request it flag values outside normal ranges and generate questions for your doctor

Why it matters: Most patients lack a unified view of their health history. AI can synthesize scattered records into actionable insights, helping you advocate more effectively during appointments and catch patterns doctors might miss across fragmented visits.

🧠 AI for Kids Tip: Build a Peekaboo app with real machine learning

MIT App Inventor's PICaboo teaches kids ages 8-12 to train an image classifier and build a working Peekaboo-style mobile app. Children learn real ML concepts through playful, supervised creation.

Setup and guardrails:

  • Age range: 8-12 years, with parent supervision recommended for first-time users
  • Access: Free at appinventor.mit.edu, works in web browsers (no downloads required)
  • Safety: All processing happens locally; no data is uploaded or shared
  • Time investment: 30-60 minutes for first complete project

Why it matters: Kids learn foundational AI concepts like training data, classification, and model accuracy through tangible creation rather than abstract theory, building confidence for future technical learning.

Cartoon 01-21-2026

Got questions about this week's updates? Hit reply and let me know what you're experimenting with.

-Pierre

Keep Reading