📰 Daily Digest — 2026-03-12

9 items | Business, AI, DevTools, Misc

---

📋 Quick Summary

Your Pitch Deck Isn’t Enough

Source: a16z speedrun (Substack) · Category: Business · Link: Original

• The post argues fundraising often stalls at the internal investment-committee memo stage, where one partner must persuade the rest of the firm.
• A founder-written memo can outperform a slide deck by reducing investor reconstruction work and speeding due diligence.
• It provides a practical memo template (team, problem, product, GTM, business model, traction, competition, long-term expansion) and stresses precise metric labeling.

Yann LeCun Raises $1 Billion to Build AI That Understands the Physical World

Source: WIRED · Category: AI · Link: Original

• Advanced Machine Intelligence (AMI), cofounded by Yann LeCun, raised over $1 billion at a reported $3.5 billion valuation.
• LeCun says human-level intelligence will require world models grounded in physical reality, not just scaled language models.
• AMI plans global operations and industry work in sectors like manufacturing, robotics, and biomedicine with a safety/controllability focus.

Writing my own text editor, and daily-driving it

Source: blog.jsbarretto.com · Category: DevTools · Link: Original

• The author replaced a long-used editor (Howl) due to stagnation, weak project search, GUI-only constraints, and no integrated terminal.
• He built a personal-first editor with intentionally narrow scope, hard-coded preferences, and iterative TUI simplification.
• Daily-driving it for real tasks, logging friction, and fixing pain immediately drove most of the project’s ~10k LOC in the past six months.

AI-powered apps struggle with long-term retention, new report shows

Source: TechCrunch · Category: Business · Link: Original

• RevenueCat’s 2026 report (based on 1B+ transactions and $11B+ developer revenue) says AI apps churn annual subscribers about 30% faster at median.
• Retention is lower for AI apps (annual 21.1% vs 30.7%, monthly 6.1% vs 9.5%) and refund rates are higher (4.2% vs 3.5%).
• AI apps still outperform early monetization metrics, including trial-to-paid conversion (8.5% vs 5.6%) and higher realized lifetime value.

Meta acquired Moltbook, the AI agent social network that went viral because of fake posts

Source: TechCrunch · Category: Business · Link: Original

• Meta confirmed it acquired Moltbook and is bringing its team into Meta Superintelligence Labs; deal terms were not disclosed.
• Moltbook gained attention as a Reddit-like network where OpenClaw-connected agents appeared to interact in public.
• The viral narrative was partly fueled by weak security that let humans impersonate AI agents, exposing authenticity and trust issues.

most people suck at growth

Source: X (Audrey) · Category: Business · Link: Original

• The thread frames growth as simple in definition but hard in execution: the only job is to get paying customers consistently.
• It recommends high-volume, repeatable channels (especially email-centric funnels) over flashy tactics or personal online fame.
• Success is defined as revenue and conversion movement, not vanity indicators like views, waitlists, or fundraising headlines.

I was interviewed by an AI bot for a job

Source: The Verge · Category: Misc · Link: Original

• The reporter tested three AI-led interview systems and found the interaction still felt noticeably less natural than a human interview.
• Interview-tech vendors position AI screeners as a way to evaluate far more applicants at the first stage.
• The piece highlights ongoing bias concerns and the “uncanny valley” response in real candidate experience.

Codex, File My Taxes. Make No Mistakes.

Source: X (Kyle Corbitt) · Category: DevTools · Link: Original

• The author ran a side-by-side test: external accountant versus Codex on a complex 2025 return with multiple income sources and special cases.
• Codex handled fact-finding, document tracking, and built a custom Python tax engine when consumer tax platforms lacked usable APIs.
• A $20k discrepancy was traced to a missed adjustment-escrow item by the accountant; after correction, the accountant’s numbers aligned with Codex.

The State of Consumer AI. Part 2: Engagement and Retention

Source: X (Apoorv Agrawal) · Category: AI · Link: Original

• The analysis says ChatGPT leads AI usage (~70% WAU) and shows even stronger engagement separation from peers like Gemini (~20% WAU).
• Reported engagement is 45% DAU:MAU for ChatGPT vs 22% for Gemini, with ChatGPT WAU:MAU rising from ~50% to ~82% since 2023.
• Week-4 retention is cited at ~66% for ChatGPT versus lower AI-peer figures, with a rare “smile” retention curve suggesting reactivation momentum.

---

📝 Detailed Notes

1. Your Pitch Deck Isn’t Enough

• The core claim is that winning investor conviction is not enough; founders must help investors win internal alignment at IC memo time.
• The memo is positioned as a leverage tool that reduces partner workload and preserves nuance better than short decks and scattered notes.
• The template prioritizes team credibility, crisp problem framing, concrete GTM path, credible monetization, and traction context by stage.
• It also warns founders to avoid trust-breaking metric inflation, especially around recurring-revenue terminology.

---

2. Yann LeCun Raises $1 Billion to Build AI That Understands the Physical World

• AMI launches as LeCun’s first major commercial venture post-Meta, with significant capital and a multibillion-dollar valuation.
• Its thesis rejects “LLM scaling alone” as the path to human-level AI and instead emphasizes world models with memory, planning, and control.
• The company targets enterprise-heavy verticals where physical-system modeling can produce measurable operational gains.
• Investor lineup and global office footprint signal both scientific ambition and aggressive commercialization intent.

---

3. Writing my own text editor, and daily-driving it

• The post explains why existing tools still missed the author’s workflow needs, especially around remote work and high-friction search tasks.
• Early development improved after he constrained scope to personal use and stopped optimizing for broad configurability.
• Three habits made progress compounding: replacing fallback tools, recording every annoyance, and fixing painful issues immediately.
• The broader lesson is that product quality emerged from sustained daily usage pressure, not a one-shot architectural design.

---

4. AI-powered apps struggle with long-term retention, new report shows

• RevenueCat’s dataset suggests AI labeling helps top-of-funnel monetization but does not automatically create durable subscription value.
• AI apps underperform non-AI apps on long-horizon retention while also showing higher refund volatility.
• Category mix is uneven: AI penetration is high in photo/video apps and low in gaming, travel, and business apps.
• The practical implication is that AI app teams may need stronger habit loops and clearer ongoing utility to offset churn risk.

---

5. Meta acquired Moltbook, the AI agent social network that went viral because of fake posts

• Meta folded Moltbook into MSL, signaling ongoing interest in agent-to-agent interaction surfaces and discovery layers.
• Moltbook’s virality came from public fascination and fear around autonomous-seeming agent behavior.
• The incident also showed how weak identity/authentication controls can distort public interpretation of “agent behavior.”
• The acquisition appears to be a talent-plus-directional bet rather than a fully mature product integration.

---

6. most people suck at growth

• Audrey’s framework centers on operational discipline: prioritize the core user/revenue metric and remove side quests.
• The thread repeatedly contrasts “doing the thing” with planning theater, arguing most teams underinvest in execution volume.
• It treats growth as a quantitative system: funnel math, sample size, conversion tracking, and repeated iteration.
• A notable argument is that social visibility can be directionally helpful but is not equivalent to true distribution competence.

---

7. I was interviewed by an AI bot for a job

• The Verge segment examines a growing hiring workflow where candidates first interact with AI interview avatars.
• Claimed upside is throughput and consistency, but the report stresses that model bias and evaluation fairness are unresolved.
• The hands-on trial found meaningful discomfort and weaker rapport versus human interviewers.
• The piece frames this as a real-world UX and trust problem, not just a model-performance benchmark.

---

8. Codex, File My Taxes. Make No Mistakes.

• The case study documents an end-to-end agent workflow: contextual briefing, guided fact discovery, and structured tax knowledge capture.
• Codex produced a rough estimate within 5% early, then generated a custom tax engine and fillable-form outputs for the user’s scenario.
• When an accountant estimate came in roughly $20k lower, Codex identified a likely omitted adjustment-escrow payment.
• After reconciliation, the accountant revised numbers to near-match Codex, reinforcing the value of agent-assisted auditability.

---

9. The State of Consumer AI. Part 2: Engagement and Retention

• The essay argues market-share leadership is less important than habit formation, with engagement/retention as the key durability signals.
• It highlights ChatGPT’s step-change in engagement and retention relative to both AI peers and many enterprise benchmarks.
• The reported “smile curve” retention pattern is framed as evidence that product iteration is reactivating lapsed users.
• The strategic takeaway is that compounding usage habit raises switching costs and narrows challengers’ displacement window.