In partnership with
FROM OUR PARTNER
200 Ways To Make Money With AI
Ready to transform artificial intelligence from a buzzword into your personal revenue generator?
HubSpot’s groundbreaking guide "200+ AI-Powered Income Ideas" is your gateway to financial innovation in the digital age.
Inside you'll discover:
A curated collection of 200+ profitable opportunities spanning content creation, e-commerce, gaming, and emerging digital markets, each vetted for real-world potential
Step-by-step implementation guides designed for beginners, making AI accessible regardless of your technical background
Cutting-edge strategies aligned with current market trends, ensuring your ventures stay ahead of the curve
Download your guide today and unlock a future where artificial intelligence powers your success. Your next income stream is waiting.
TODAY AI BRIEF
3 things that happened while you were busy
1 Anthropic published what AI did to its own hiring, and juniors lost.
The share of Anthropic's technical hires who were junior fell from 27% to 9% since early 2025, and junior staff dropped from 21% to 14% of its technical departments, with job postings for them falling sharply after December 2025. The more revealing number is the idle one: about 80% of the time in an interactive Claude session is spent waiting for a human to respond. This is Anthropic's own internal data about itself, with no outside audit. Read Anthropic's report.
2 Meta shipped a standalone app in eight weeks by refusing to call code done.
Meta built Forum, a separate app for Facebook Groups, with a small team and agents doing much of the work, reaching core feature parity in eight weeks. The rule that made it work was narrow: a change counted only once it built, passed tests and survived review, so compiling was never enough. Stress tests that hammered the app found a memory problem the agent then reproduced and fixed, roughly halving memory use, and an engineer still signed off on every shipped change. The figures are Meta's own, published without baselines. Read Meta's engineering post.
3 Anthropic opened up the review system behind 30,000 code changes a day.
Several agents hunt for bugs in parallel, a separate group votes on each finding, and only confirmed ones get posted, sorted by severity and tagged so inherited bugs are not blamed on whoever touched the file last. Human approval is then scaled by risk rather than by seniority: none for low-risk changes, one person for ordinary ones, two for the dangerous paths. An agent's approval never counts as a human's, and approvals expire when new commits land. Substantive review comments rose from 16% to 54% of internal pull requests after it launched. Read the full playbook.
Bonus reads: Claude Code's hooks guide, which is today's tutorial and the cheapest version of Meta's rule, plus a clear-headed account of the OpenCode key breach if you store provider keys in any third-party console.
AI FORWORD
The code is nearly free now. The queue is a person reading it.
The idle machine. The most quietly damning figure in Anthropic's report on itself is that roughly 80% of interactive session time is spent waiting on a human. The expensive model is not the constraint. The person who has to look at the output is.
Who gets squeezed. That bottleneck has a shape in the hiring data. Junior technical hires fell from 27% to 9% of the intake, because the work juniors used to do is the work agents now do, while the reviewing that remains is exactly the work juniors are not yet trusted with. The ladder is losing its bottom rung.
The engineering answer. Meta's response was to move the proof earlier, defining done as built, tested and reviewed rather than written, and attaching the evidence to every change. That turns review from reading code into checking whether the evidence holds up, which is a faster and more teachable job.
The organizational answer. Anthropic went further and rationed human attention deliberately. Reviews are capped at three rounds before someone rethinks the approach, which took changes needing five or more rounds from 8.9% down to 1.5%. The detail worth stealing: about 59% of what its reviewer kept flagging was code that predated the review, so the endless loops were mostly rediscovering old bugs.
The reframe. Every one of these fixes buys back the same scarce resource, which is a qualified person's attention, and none of them creates more qualified people. That is the gap worth watching, and the opportunity in it. If you are early in a career, the valuable skill is no longer writing the thing. It is being the person trusted to say whether the thing is right.
PRESENTED BY :
100+ Claude Code hacks to ship code 10X faster
Top engineers at Anthropic say AI now writes 100% of their code.
Are you using AI to write yours?
These 100+ Claude Code hacks show you exactly how. Sign up for The Code and get:
100+ Claude Code hacks — free
The Code newsletter — learn the latest AI tools and skills to code faster in 5 mins a day
SOCIAL SNAPSHOT
What people are actually watching and sharing
Rotate your keys. OpenCode says a breach of a system it was decommissioning may have exposed provider keys for 1.4% of Console users, and it is telling them to revoke and replace those keys now. Dollar figures circulating online come from user posts, not from OpenCode or the providers.
Fifty rounds overnight. An agent author and an agent reviewer can argue with each other forever. One Anthropic pull request went through more than 50 review rounds in a night before anyone noticed, which is why the round cap exists.
Robots that use the app like people. Among Meta's checks were navigation agents that drive the live app the way a user would, running the same paths repeatedly to catch regressions that unit tests never see.
A terminal setup worth copying. A senior developer walks through building a terminal setup you actually stick with, combining a fast terminal, session management, fuzzy search and a coding agent into one workspace.
JavaScript as a neural net. Handwired turns ordinary conditionals and loops into matrix maths that produce the same answers, with no training involved, so the resulting program stays fully differentiable. It is an experiment rather than a tool, and all the more interesting for it.
PROMPT STATION
Make any AI prove it finished before you believe it
The thread running through today's issue is that finished should mean verified, and that rule is not only for code. This prompt works in Claude, ChatGPT or Gemini and forces the AI to define its own finish line before it starts, then show its evidence against that line at the end. Use it on anything where being quietly wrong would cost you.
Before you start [THE TASK], write a DONE CHECKLIST: the specific, checkable conditions that would prove this was done properly. Make them things that can be verified, not opinions. Include the one condition most likely to fail.
Wait for me to approve the checklist. Do not begin the work until I reply.
Once I approve, do the work. Then go back through the checklist item by item and mark each one met, not met, or cannot be checked, with the specific evidence from your own output for each. Quote the evidence, do not describe it.
If any item is not met, say so at the top in one sentence before showing me anything else. Never mark an item met because it probably is.