Your heart occasionally skips a beat without you noticing.

It’s normal, it happens, and most people don’t even feel it.

So no, babe, not every skipped beat is your crush. Sometimes your heart just wants down time.

What’s in store:

Read Time: 5 minutes

OPENAI

Small, quiet and silently destructive

OpenAI’s Codex CLI may be putting huge strain on some users’ SSDs because of a logging bug that is still open on GitHub.

The issue was flagged by GitHub user 1996fanrui on June 14, after they noticed unusually high disk activity.

They found that Codex was constantly writing diagnostic logs to a local SQLite database stored at ~/.codex/logs_2.sqlite.

Over 21 days, their drive recorded around 37TB of writes. If that continued for a full year, it would add up to about 640TB.

For context, some 1TB consumer SSDs are rated for around 600TBW across their lifetime, meaning this bug could burn through a drive’s warranted endurance in under a year.

Very normal behaviour from a log file... that is apparently auditioning for a demolition job.

The problem seems to come from a logging setting that was likely never meant for regular users.

Codex’s SQLite feedback sink runs at global TRACE level by default, which is the most detailed logging mode.

That means it records a huge amount of low-level activity, including WebSocket payloads and routine file system events.

What makes it worse is there’s no way to reduce the level of logging - it’s hardcoded to always do the absolute maximum.

The database file may not look huge, but the constant insert-and-delete activity can still cause far more physical writes than the file size suggests.

The main points:

Logs gone ferocious

For now, Linux and macOS users can redirect ~/.codex/logs_2.sqlite to /tmp/ using a symlink, which moves the writes to temporary storage.

The file does not contain conversation data, so losing it after a reboot should be fine.

Tiny silver lining: your SSD is not dying in battle. It is being slowly defeated by admin.

Codex logs every time you blink. - MV

Vote for live results + see results and opinions from yesterday at the bottom of the email.

AI PROMPT PACK

1,000 Copy-Paste AI Prompts Swiped From The Experts

This cookbook-style database is fully searchable, filterable, and color-coded for tracking down primo prompts in seconds. 

See all 1,000 AI use case recipe cards to save time, make profits, and 10X your business’ potential in 2026. 

Try our AI Cookbook: 

Number Crunch

Two angles in a triangle are 42° and 73°. What's the third angle?

Find the answer at the bottom of the email!

Watch This

Math gets a lot easier once it stops feeling like punishment. In this video, Han breaks down why maths can feel so frustrating at first, then shares a more practical way to study that makes problem-solving feel a bit less miserable and a lot more effective.

ANTHROPIC

The office memory bank

Anthropic is bringing Claude further into Slack with Claude Tag, a new feature that lets teams tag @Claude in channels for help, insights and task support.

It builds on Claude’s existing Slack tools, but adds more context.

Claude can follow a channel’s work, remember what has been discussed and, if allowed, pull useful information from other parts of the company.

In brief:

No more office amnesia

Each channel can have its own Claude, with admins deciding what it can access.

That means legal Claude does not casually wander into engineering like an intern with no badge.

Claude Tag can break tasks into steps, post updates in Slack threads and even proactively flag forgotten work or relevant updates.

The bigger picture is simple: enterprise AI is moving from one-off chatbot replies to tools that understand company context.

Microsoft, Snowflake, Databricks and Glean are all chasing the same prize.

The feature is useful coz most workplace knowledge currently lives in Slack messages, half-finished docs and one person named James who is always on annual leave.- MG

Mindstream Picks

Health: France has confirmed its first Ebola case linked to the outbreak in Congo, with health officials stressing the risk to the wider public remains very low.

Gaming: Rockstar may still be planning a Grand Theft Auto 6 Collector’s Edition, despite only standard and Ultimate editions being revealed so far.

Movies: Al Pacino’s new Netflix crime drama In the Hand of Dante has debuted to a disappointing critical reception, earning a low Rotten Tomatoes score.

Don’t Miss: Amazon-owned Zoox has revealed an updated version of its purpose-built robotaxi, featuring a more comfortable interior and refinements aimed at preparing the vehicle for mass production. The company is currently testing its autonomous service in several US cities while awaiting regulatory approval to launch commercially.

AI Art

Image of the Day

Daily Image Prompt

Submit your artwork to Mindstream →

Yesterday’s Poll

“Can AI actually make open source software safer?”

Yes, defenders need better tools - 77% ✅

No, this sounds like chaos with branding - 23%

Your Views:

“Just waiting till one of these companies exploit the bug to violate the privacy of the users and then go 'ooops, we did it again'” - surya

“Security is a game of thief and police, the police needs to stay ahead. so lets the 'powers of the AI' help them” - info

“It looks like the AI will learn every posdible Open Source of the Planet. Fix the Bugs but at the same time put extra randomness code for future execution. OpenAI want to Brand itself and wants learn and update iteself creating distribution itself on the Planet.” - jaketho

Submit your opinions in our polls to be featured!