Welcome back. By 2030, AI could write an internet’s worth of text every single day, one startup predicts. So how will you find useful and relevant information from the pile of AI slop? Today’s top story covers a new search model built for that flood. Meanwhile, GPT-6.2 Astra just landed, and devs are already divided.

Also: Why devs remain valuable in the AI coding era, explore a Claude Code demo highlighting how devs build for Apple Duo, and learn how to stop Windows 11 from secretly tracking your workflow. 

Today’s Brief

  • A senior engineer’s advice on what to build in 2027

  • Jensen Huang’s emotional statement and the board’s response

  • Turn Hermes into an autonomous software factory (tutorial)

  • How to verify coding agent work at scale (cookbook)

TODAY IN PROGRAMMING

Click here to check out how SID. ai’s new models work.

A new Silicon Valley startup beats Opus and Astra at search: Bigger AI models keep getting better at math and coding, but making them larger has barely improved web search. San Francisco startup SID is taking a different route. Instead of relying on one massive, expensive model, its new SID-2 system uses teams of up to dozens of agents working together on complex queries. As a result, SID claims its system beats frontier models like Opus and Astra while running 8x faster and cutting search costs by 20x. It prices out at just $0.25 to $2 per million tokens. Give them a spin here.

OpenAI’s cheaper coding models divide devs: The ChatGPT maker just dropped GPT-6.2 Astra and Sol in Codex and the API, with both models promising to get the job done using fewer output tokens. Astra also gets a 20% price cut to $8/$40 per million input/output tokens. But early reviews are mixed. Some devs are seeing cleaner code and better reviews, while a tech CTO says Astra gets stuck on long tasks and asks baffling questions. You can start trying it with the prompting guide and decide for yourself.

Apple outage locks devs out of their own systems: An internal update went wrong at the iPhone maker, leaving employees locked out of sign-ins, VPNs, and internal builds for hours. The App Store and iCloud were hit too, and even Apple’s own status page showed green for hours while users piled onto Downdetector. Coming just a week after Microsoft’s outage, it’s another reminder to stage your rollouts and keep status tools off shared infrastructure. Apple says it wasn’t a cyberattack. More details here.

AI lets teams ship code faster than ever. The same acceleration applies to risk. AI coding assistants create new attack surfaces most legacy tools miss: prompt data leakage, hallucinated packages, and overprivileged agentic workflows.

This new cheat sheet covers the five unique risk surfaces of tools like Claude Code and gives practical, day-one guidance for each. It walks through prompt hygiene, secure configuration, catching slopsquatted packages, secrets management, and CI/CD guardrails.

Start securing AI-assisted development today.

INSIGHT

A senior engineer’s advice on what to build in 2027

Source: The Code, Superhuman

What’s next? As teams start planning for 2027, Bjorn Roche, who manages engineering teams at Rutter and formerly worked at GIPHY, calls 2026 the year the industry got a glimpse of what AI can do. Agents are already writing most of the code at many companies, but knowing what to build is a whole different challenge.

Bolt it onto what works. Roche’s go-to move for figuring that out is to add AI to a product that’s already working. Or to take a proven human workflow and let AI handle one step at a time. That gives the team a baseline to improve on.

Here’s the rest of the cheat sheet:

  • Old ideas, new math. Hunt for workflows that were once too labor-heavy to automate, then add AI to the human version. Rutter did that when third-party support bots couldn’t handle its technical questions. The team built AI drafting into its own support flow and kept humans in the loop.

  • No wand included. AI will not make every project easy, so you should plan major migrations as labor-intensive efforts and take ownership of the decision to patch or rewrite. 

  • Skip the wireframe. Have AI build a working prototype instead. It follows existing design styles well enough to match wireframe speed, and a prototype is easier to evaluate. Leave branding and polish to designers.

  • Slop stays home. Read and trim AI output before it reaches a teammate. If AI is skipping decisions you should make or creating work for someone else, you’re using it wrong.

The review queue. When agents write more code, you’ve got more PRs to review. While AI speeds up some parts of engineering, others aren’t keeping up. Roche’s advice is to basically shift your team’s time and resources to wherever the work is piling up next. You can read the full article for more of his advice.

PRESENTED BY MOZILLA FOUNDATION

Or an idea that’s too early, unexpected, or ambitious for VC funding? Mozilla Foundation wants to back you. Every selected project comes with:

  • Funding

  • Expert coaching

  • A global network of peers

  • Showcase opportunities, like Mozilla Festival

IN THE KNOW

What’s trending on socials and headlines

Meme of the day.

  • Raised Ceiling: AI may write most code, but a SpaceXAI engineer explains why the best software will still come from people who care, and she has the perfect analogy (2K likes).

  • Jensen’s Plea: NVIDIA’s ousted founder broke his silence in an emotional open statement. It closes with an offer the board has reportedly already answered (45.6M views).

  • App Origami: Devs are already testing their apps on Apple’s foldable iPhone Duo without ever touching one. The way Claude Code pulls it off is oddly satisfying to watch (1M views).

  • Windows Watching: A researcher found Windows 11 quietly logging which AI coding agents you run and uploading it to Microsoft. The way it spots them is almost comically simple (3.6K likes).

  • Autonomous Attacks: Google’s threat team found attacks that exploit fresh bugs faster than any team can patch, entirely without humans. The full report shows how they pick targets (4.2M views).

  • Brain Drain: An engineering manager’s brutal self-audit of what agents did to his skills is making the rounds. Symptom four already has a nickname (3.1K likes).

  • No Coincidence: Palo Alto Networks’ CEO says this week’s cyberattacks are connected, and the worst part is there’s no silver bullet. His advice on what to do before you’re next has devs talking (2.5K likes). 

TOP & TRENDING RESOURCES

Click here to watch the tutorial.

Top Tutorial

How to turn Hermes into an autonomous software factory: This tutorial shows how to set up Hermes to watch GitHub issues, write specs, build and validate fixes, and open pull requests with minimal intervention. You’ll learn how to install the factory skill, connect GitHub, run the first issue end to end, and schedule the whole workflow to keep working across repos.

Top Tool

Rapid Router: Use Claude, GPT, Gemini, DeepSeek, and other major models through one API key and balance. Point your existing SDK at it and each request is automatically routed to the cheapest healthy provider, with failover built in and live latency, uptime, metering, and spend tracking handled for you.

Trending Cookbook

How to verify coding agent work at scale (by Anthropic): This guide breaks down the verification system behind Claude Code, where ~90% of the code is written and reviewed by Claude. You’ll learn which tests are actually worth investing in, how to build fast verification loops agents can trust, and how to turn misses into rules so you can delegate more without manually checking every diff.

AI CODING HACK

How to find out if your new tech stack is any good

A new library drops and you lose an afternoon reading docs, only to guess whether it fits your codebase.

A senior dev shared a prompt that makes your coding agent answer with evidence instead: a working proof of concept in a draft PR, graded and verified. He demoed it on the XState v6 beta.

  • Step 1: Open Claude Code in the repo where you’d actually use the tool.

  • Step 2: Paste the full prompt, swapping the first line for the technology you’re vetting. It starts like this:

Let's assess [TECHNOLOGY].

1. Research what it does and the problem it solves.
2. Find the gnarliest problem in our application that it might assist with
3. Create a proof-of-concept implementation in a draft PR
...
  • Step 3: Review the draft PR and the grades on API nuance, typesafety, testability, complexity, docs, and deployment changes.

  • Step 4: If the recommendation is skip, delete the branch and move on.

Whenever you're ready to dive deeper

If you enjoyed this newsletter, you should definitely check out our IT and cybersecurity newsletters. Subscribe using the links below:

We put together 20+ hands-on guides covering coding agents, agentic engineering, and leadership frameworks to help you level up your career. You can get them here.

Advertise with us: Join companies like Google, IBM, and Datadog. Showcase your product to our 400K+ engineers and 150K+ followers on socials. Get in touch.

What did you think of today's newsletter?

Your feedback helps us create better emails for you!

Login or Subscribe to participate

You can also reply directly to this email if you have suggestions, feedback, or questions.

Until next time — The Code team