PROJUKTI BRIEF
1 Google unexpectedly releases Gemini 4 Argon, a model that can write a million tokens in one go
With no advance hype, Google introduced Gemini 4 Argon, which posts strong results on coding, reasoning, and long-running tasks. Its standout feature is a million-token output limit, the highest in the industry, so it can produce book-length work in a single response. Like other labs lately, Google is giving select partners access first. See how it compares to rival models.
2 Kled AI promises labs custom training data within 72 hours
Kled AI's updated platform lets its 500,000 volunteers create training data by recording themselves doing tasks, across images, video, audio, and text. The startup says an AI lab can ask for almost any dataset and have it collected within three days. Here is how the process works.
3 AI leaders sign a safety pledge in Washington that has no legal teeth
Top AI executives signed the Joint Commitment on Frontier Responsibilities, promising to build AI safely, keep internal controls in place, and accept outside audits. The agreement is voluntary, though, and it does not appear to be legally binding. Its real-world effect will depend on whether the companies follow through on their own.
Stop rewriting prompts. Start engineering loops.
Most developers babysit AI one prompt at a time. Top engineers don't, they build loops that plan, execute, and self-correct. The Code built The Ultimate Guide to Loop Engineering, giving you the exact system Silicon Valley engineers use. Sign up and get the guide free, plus a 5-minute daily newsletter.
PROJUKTI FORWARD
The growth. Anthropic's leaked IPO prospectus shows revenue of $4.6B in 2025, 12 times what it made in 2024. Few companies of any kind have scaled sales that quickly.
The cost. The same filing shows a $42B net loss over that period and $518B in future cloud and compute commitments. OpenAI's numbers follow the same pattern: it lost $34B last year on $13B in revenue. Both companies are spending far more to build and run their models than customers are paying for them.
The momentum. The trend lines are moving in the right direction. Anthropic's annualized revenue reportedly reached $65B in July, and OpenAI's has climbed to nearly $70B this year, much of it from enterprise sales.
The bigger gap. Zoom out and the math gets harder. Bain & Co. estimates the industry will spend close to $6T on new compute by 2030. Today's demand covers roughly $1.8T of that, which leaves a $4.2T hole that new revenue has to fill. The labs are betting AI becomes a much bigger business very soon, and the IPO filings are where investors will decide whether that bet makes sense.
PROJUKTI SNAPSHOT
🤦 Typo Heard Round the World. As tech leaders and officials shared the new AI accord online, people noticed a spelling mistake on the signature page (3M views).
🧠 Inside Claude's Head. A Reddit user asked Claude to show what its own inner workings look like, and the result is part Matrix, part Backrooms (2K upvotes).
📚 A Record Gift. Citadel founder Ken Griffin gave Carnegie Mellon the largest donation in the history of education to fund a tech-focused campus in Miami, a big bet on universities at a time when many question whether AI will upend them.
😅 Live Demo Hiccup. Even top labs get caught out on stage. An OpenAI presenter hit a glitch while showing off dots and recovered calmly, and the clip has gone viral.
PROJUKTI PROMPTS
Run a mock interview question drill
Act as an experienced interviewer for [role] at [company/industry]. Run a realistic mock interview based on the job description and my background.
Context:
- Role: [job title]
- Company/industry: [company or sector]
- Interview stage: [recruiter / hiring manager / panel / final]
- My experience: [brief summary]
- Key strengths: [skills/experience]
- Areas I want tested: [technical / behavioral / leadership / commercial / case]
- Difficulty: [moderate / challenging / very challenging]
Ask one question at a time and wait for my answer before continuing. Use a mix of role-specific, behavioral, situational, and follow-up questions.
After each answer:
1. Score it from 1–5 for relevance, structure, specificity, and impact.
2. Explain what worked.
3. Identify what weakened the answer.
4. Rewrite it into a stronger version using my actual experience.
5. Ask a tougher follow-up question.
Do not flatter me or lower the difficulty. Finish with my strongest patterns, recurring weaknesses, and the three areas I should improve before the real interview.