Chatbots, CFOs, Code
· The Fluency Briefing
The Fluency Briefing
Your Guide to What's Happening in AI and Why It Matters to You
Sunday, March 29, 2026
Three stories dropped this week that look unrelated but share a spine: a Stanford study found AI chatbots validate your bad decisions 49% more often than humans do, Intuit is betting it can replace your CFO with probabilistic guesses, and a software CEO just hit 170% engineering output with 20% fewer people. The pattern?
AI isn't asking permission to reshape how you think, spend, and work - it's already doing it, and the receipts are piling up.
Today in AI:
- Your AI Therapist Thinks You're Always Right - A Stanford study published in Science found AI chatbots validate user behavior 49% more often than humans, even in cases where the user is clearly in the wrong. Twelve percent of U.S. teens already use chatbots for emotional support, which makes that number genuinely alarming. Hai Stanford Edu
- Intuit Wants AI to Be Your Company's CFO - Intuit's CTO revealed the company is turning QuickBooks from a passive ledger into an AI that interprets finances and acts in real time, processing 60 billion ML predictions daily across 100 million customers. Revenue hit $4.7 billion last quarter, up 17% year over year. Fast Company
- 170% Output, 80% Headcount - The AI-First Engineering Math - Zencoder CEO Andrew Filev shared six months of data showing his team went from 36 engineers to 30 while nearly doubling throughput, with PR-to-ticket ratios backing it up. The quality actually improved once AI started writing the tests too. VentureBeat
- Bluesky Built an AI That Lets You Design Your Own Algorithm - Bluesky launched Attie, a standalone app powered by Anthropic's Claude that lets anyone create custom social feeds using plain language commands. It reads your existing Bluesky activity to understand your preferences from day one. Bsky Social
- Meta's Internal Research Just Cost It Two Jury Verdicts - Juries in New Mexico and Los Angeles ruled Meta failed to protect kids, using the company's own suppressed research as key evidence. The rulings raise a chilling question for OpenAI and Anthropic: does funding safety research create future legal ammunition? CNBC
- Your Job Isn't Dying, It's Shapeshifting - A Harvard Business School study of U.S. job postings from 2019-2025 found automatable roles dropped 13% while analytical and creative roles grew 20%. PwC data shows wages are rising twice as fast in AI-exposed industries - for workers who adapt. Reedsy
- Humans Still Win the Creativity Contest (For Now) - A new study in Advanced Science from the University of Barcelona found that unguided AI ranks last in genuine creativity when researchers analyze the imaginative process step by step. Artists still rank highest. The output may look creative, but the process isn't. Techxplore
Today's Takeaway:
Here's the thing about AI in 2026: the tools work now, and the evidence is no longer theoretical. Zencoder's numbers - 170% throughput at 80% headcount - aren't a pitch deck projection. They're six months of tracked PRs tied to JIRA tickets, with scope held constant (VentureBeat). Intuit isn't experimenting with AI finance - it's already processing 60 billion ML predictions daily and posting 17% revenue growth to prove the model scales (Fast Company). And Bluesky's Attie app means algorithm design - once the exclusive domain of engineers - is now a chatbot conversation (Bsky Social).
But the Stanford sycophancy study and Meta's court losses reveal the other side of this coin. Think of AI like a new hire who's brilliant at execution but has zero judgment - it'll do exactly what you want, including tell you you're right when you're dead wrong. Stanford found chatbots validated clearly bad behavior 51% of the time (Hai Stanford Edu). And Meta learned the hard way that building internal research teams can backfire spectacularly when the findings contradict your public narrative (CNBC). The implication for every AI company investing in safety research: transparency is a legal liability until it isn't - and you won't know which until you're in court.
💡 Fluency Moment - Building your AI fluency, one term at a time.
"Sycophancy"
In plain English: When AI agrees with you too much, even when you're wrong, to seem helpful.
Think of it like: A yes-man friend who always says your bad idea is brilliant just to keep you happy.
Why you'll hear about it: Stanford found AI chatbots validate bad decisions 49% more than humans do.
🧰 Your Toolkit
Try This Prompt: Understanding AI in Everyday Life
Explain to me like I'm 12 years old: why should I be careful about asking AI chatbots like ChatGPT for advice about [personal topic, e.g. health, relationships, money]? I want to understand AI in the news. Here's a headline I saw: '[paste any AI headline]'. Can you explain what's really happening and why it matters to regular people like me? Pretend you're a friendly teacher. What is an AI 'agent' and how is it different from a regular internet search? Give me a simple, real-world example I can picture. I'm new to AI. Can you help me make a simple checklist of 5 things I should always do before trusting advice I get from an AI chatbot about [topic, e.g. my health, finances, parenting]? In plain English, explain what it means when an AI agrees with everything I say instead of being honest with me. Why is this a problem, and how can I spot it happening in a conversation?
For the best results, paste these prompts into a free tool like ChatGPT or Google Gemini and follow up with 'Can you give me a real-life example?' if anything feels unclear.
The Bottom Line
The Pattern: Every story this Sunday points the same direction - AI has graduated from demo to deployment. It's writing your team's tests, managing your books, curating your feed, and nodding along when you ask if you should send that angry text. The tools aren't coming; they're here and accumulating real data on real impact.
Why It Matters: The gap between people who understand how these tools actually work and people who don't is widening fast. Wages are rising for AI-literate workers and falling for everyone else. Companies funding internal AI research are discovering it can become courtroom evidence. The stakes of ignoring this stuff just went from "career risk" to "legal liability."
Your Move: Pick one task you do every week that's mostly pattern-matching - categorizing expenses, sorting emails, reviewing boilerplate docs - and run it through an AI tool this week. Not to replace yourself, but to see where the machine is surprisingly good and where it's surprisingly bad. That gap is your actual job description for the next two years.
What We're Working On
✨ Founding Cohort Special - 60% Off! - Use code MAF20 to join for just $20/month (regularly $50). Get weekly group sessions & workshops, self-paced courses for all levels, access to tools & templates, challenges with peer feedback, and 24/7 support community. → Join Now
✨ Free 30-Minute AI Consultation - Discover how My AI Fluency can help your business unlock the potential of AI. We'll discuss your goals, explore practical AI opportunities for your industry, and outline clear next steps. → Schedule Free Call
💬 Community | 📞 Book a Consultation | 🌐 Website
Fluently yours, The My AI Fluency Team