🚀 DeepSeek Levels Up, AI Crushes Black Friday, and 11.7% of Jobs Are in Play 💼
ATH
Tue, Dec 2, 2025
Welcome
This week, AI felt incredibly practical.
Open-source models are closing in on GPT-5, holiday shopping numbers showed what happens when AI sits at the checkout, and fresh research from MIT says nearly 12% of U.S. jobs could be done by today’s AI systems.
On the builder side, Claude’s coding mode still scares beginners more than it should, while consulting giants quietly deploy thousands of AI agents behind the scenes.
Let’s translate all that noise into things you can actually use.
AI News That Matters
🐉 DeepSeek Strikes Again – Open Source Aims at GPT-5
Chinese lab DeepSeek just rolled out V3/V3.2 models that match or beat GPT-5 and Gemini 3 Pro on several reasoning benchmarks—at 10–25× lower cost and under an MIT-style open license.
Why it matters:
- Frontier performance is no longer a closed-lab privilege.
- Builders can get “top tier” capability without proprietary pricing. Expect a wave of startups built entirely on open models.
💻 Why Beginners Are Scared of Claude Code (and How to Fix It)
A recent engineering write-up dug into why many beginners avoid Claude’s coding environment even though it’s powerful:
- Terminal + file access feels “too dangerous”
- Permission prompts interrupt flow
- People aren’t sure when to trust autonomous edits
How to get past it (quick playbook):
- Start in permission mode, not full autonomy.
- Let it edit copies of your files, not the originals.
- Use it first for small chores: test writing, refactors, docstrings.
Once confidence goes up, so does productivity.
🧱 McKinsey’s Quiet AI Agent Army
A recent analysis of one global consulting firm showed a clear pattern:
- Build a data moat – clean, well-structured internal knowledge.
- Deploy AI as assistants – search, drafting, and summarization.
- Scale into AI agents – 10,000+ internal agents now handle repetitive workflows across ops, research, and finance.
Why it matters: This is the “boring but rich” blueprint: data → assistants → agents → compounding productivity.
🛒 How AI Quietly Dominated Black Friday
Black Friday set a new online spending record at $11.8B in the U.S.
The AI angle:
- AI-driven traffic to retail sites jumped 800%+ year-over-year.
- Shoppers who arrived via an AI assistant were ~38% more likely to convert.
- One major SaaS vendor says AI agents influenced $22B in global sales during the Thanksgiving–Black Friday window.
If you run e-commerce and don’t have AI in search, recommendations, and support, you’re leaving money on the table.
🧑💼 Research Confirms: AI Is Ready to Displace, Not Just Assist
Multiple recent reports paint a consistent picture:
- MIT: 11.7% of U.S. labor could be automated with today’s AI.
- McKinsey: up to 14% of workers globally may need to change occupations by 2030 due to AI and automation.
🎯 Leaked: Personalized Ads Are Coming to ChatGPT
Developers digging through the ChatGPT Android app found references to “search ads” and an ads carousel, hinting that ad units and targeting logic are being wired in behind the scenes.
What this likely means:
Search-style ads tied to queries
Highly targeted campaigns using conversation context
A big new performance channel for brands and a UX/trust challenge for users
If you rely on ChatGPT for neutral answers, watch this space carefully.
️ 🛠️ Tools That Actually Help
1️⃣ Claude Code (Permission Mode) – Great for beginners if you constrain file access and review every diff.
2️⃣ DeepSeek V3 / V3.2 – Open-source frontier models for apps that need strong reasoning without closed-API pricing.
3️⃣ DeepSeek Coder – An open model optimized for coding, similar to GPT-Code or Claude Code mode.
4️⃣ 🧹Sweep – A top-tier AI coding plugin for JetBrains IDEs (IntelliJ, PyCharm, WebStorm). It helps you debug, refactor, and write features 2x faster — and is available free on the JetBrains Marketplace.
5️⃣ ✈️ MagicTrips – Your AI-powered trip planner. Get custom itineraries built around your interests, budget, timeline, and travel style – no more scrolling through 20 blogs.
️ 🔮 Future Watch: What’s Coming Next in AI
1️⃣ Open-Source vs. Closed Labs
DeepSeek-style releases will force everyone to rethink what “frontier model” means.
2️⃣ Ads Inside Chat Interfaces
Expect the first wave of very opinionated takes when ads hit conversational AI.
3️⃣ Agent Everywhere
From consulting giants to solo founders, AI agents will quietly handle more of the boring work each quarter.
4️⃣ Job Market Rewrites
The 11.7% automation figure is a starting point, not an end state. Policy and re-skilling will decide how painful the transition is.
️ 💡 Productivity Tips
1️⃣ Start “Tiny Agents” Inside Your Team
Instead of pitching a giant AI transformation, start with one small agent:
“Summarize daily support tickets”
“Draft weekly KPI email”
“Clean and label sales notes”
Prove value in 2-3 weeks, then expand.
2️⃣ Use AI to Pre-Test Your Offers
Before you spend on ads, have models simulate different customer personas and critique your landing page, copy, and offer. It’s not perfect—but it’s cheaper than a failed campaign.
3️⃣ Build Your Personal “Job Moat”
Map your role into tasks. Mark each as:
✅ Human-only (relationships, strategy)
🤝 Human + AI (analysis, writing)
⚠️ AI-can-do-most
Then aggressively learn tools that boost you in the 🤝 column. That’s where future-proof careers live.
️ 😂 Meme of the Week
Me: “Will AI replace my job?”
AI: “No… but I might replace your least favorite tasks first.”😅
️ 🌀 Closing Thoughts
The story of this week isn’t “AI will take all the jobs” or “AI will save us time.”
It’s this: the people and companies that adapt fastest get the upside.
Everyone else negotiates with whatever’s left.
Use the tools. Build the moat. Let AI help – before it helps your competition more.
Stay curious,
AI That Helps 🤖