Total agent recall

gm legends, happy Monday.

Weekend was busy: someone built persistent memory for Claude Code and it hit 3,000 GitHub stars in three days; Artin Bogdanov (a16z Speedrun, 15 years at Adobe) launched an audio learning tool that skips its own app entirely and drops content straight into your Spotify library; AI agents started trading crypto and Polymarket positions autonomously with 10 million actions already in the books; and one founder's pricing question turned into a real debate about what annual conversion rate actually means.

Agentmemory is an open-source persistent memory layer for Claude Code, Codex, and other coding agents built by Rohit Ghumare, Principal Product Evangelist at iii.dev, who got tired of agents losing context mid-task and built the thing himself — it reduces token consumption by up to 95% and enables 200x more tool calls before hitting context limits, with memory that's portable and cross-agent rather than locked to a single platform.

🔥 Our Take: The mental model is: agent has context, agent does the work, done. The reality is you're two hours into a session, context window full, agent has forgotten everything it learned about your codebase. Agentmemory is the fix — persistent memory underneath the agent, survives resets, works across tools. Rohit built it because he was living the problem. 3,000 GitHub stars in three days is a lot of other people saying same.

SUN-to-Spotify turns any topic, book, or question into a cinematic audio course (5–90 minutes, customizable voice) and sends it directly to your Spotify library — built by Artin Bogdanov, who spent 15 years in product design at Adobe and leading the AI BI team at Walmart Labs before founding SUN, which just completed a16z Speedrun.

🔥 Our Take: SUN makes audio courses on any topic and drops them straight into Spotify — no new app to open, no new habit to build, just content showing up in your library. Artin ran AI product at Adobe for 15 years and just came out of a16z Speedrun. His read is that people aren't building new routines around another learning tool. Whether AI-generated audio is actually good enough to sit next to everything else in your Spotify is the only real question. Worth finding out.

Fere AI deploys AI agents that research signals, set entry and exit rules, and execute trades autonomously across crypto exchanges and Polymarket prediction markets — built by Akshaya Aron and Pranav Prakash, who have been building autonomous AI systems since 2014 and deployed them at Fortune 100 companies in pharma and industrial ops before raising $1.3M seed and turning the same infrastructure on financial markets.

🔥 Our Take: AI trading tools mostly stop at the backtest. Fere AI is already running — 10 million autonomous agent actions live across Ethereum, Solana, Arbitrum, and Polymarket. Akshaya and Pranav weren't building this in fintech. They were running autonomous systems in pharmaceutical manufacturing since 2014, where you can't move fast and break things. That background is showing up in how this is built.

Ilai Szpiezak (@ilaiszp) posted with a specific data point: 42% of Pretty Prompt users were choosing annual, which made him wonder whether to cut monthly entirely and just go annual-only.

The replies pushed back. Multiple commenters argued that users need low-commitment entry points before they'll trust a product enough to commit for a year — and that removing monthly doesn't increase commitment, it increases abandonment.

The sharpest cut came from Stan Kolotinskiy: "As an end-user I'd prefer having a monthly option in any product — I'm usually not keen to get locked into a year subscription." The tension is real: your metrics say one thing, your users feel another.

Good thread if you've been staring at your own annual conversion rate and wondering if it means something.