OpenAI DevDay 2026: Cheaper GPT-6.1, Pricier $500 Pro
ALSO: Anthropic’s $42B IPO loss, AI insiders sound alarm
Welcome to Guru99 AI Report!
Top Story: OpenAI just made intelligence cheaper and access pricier on the same day, AI insiders are sounding fresh alarms, and software teams are handing more of the work to agents. This edition connects the shifts that matter most to you. Let’s dig in.
🤖 OpenAI’s DevDay: Always-On Agents, Cheaper Model, Pricier Pro
Brief Buzz:
At DevDay on September 29, OpenAI ran more than 20 product launches. The headliner is Dots — agents that keep working after you’ve closed the tab. There’s also a cheaper model and a new top-tier plan, while heavy users of the $200 tier end up with less.
- Dots run on their own cloud computer and browser, and they’re rolling out first to the Pro and Business Premium plans.
- GPT-6.1 costs $2 per million input tokens and $10 per million output tokens — OpenAI says it delivers intelligence close to Astra at one-fifth the price.
- OpenAI says Ultrafast runs up to eight times faster in Codex, and the new $500-a-month Pro 500 plan includes it.
- From October 30, $200 Pro users get only half their Codex and work allowance, and GPT-6 Pro message limits drop to 100 per week.
- Eligible companies can spend part of their OpenAI commitment on 32 partner products like Figma and Harvey through the new OpenAI Marketplace.
💡 Why Should You Care?
DevDay is squarely aimed at power users and businesses, not casual users. Unsupervised agents could save real hours — but the best features now cost more, and $200 Pro subscribers pay the same for less.
⚠️ AI Insiders Warn the Race Is ‘Gambling With Our Lives’
Brief Buzz:
AI-safety warnings are getting louder. In a single week, researchers from Anthropic and OpenAI resigned or spoke out, arguing that the race toward superintelligence is moving faster than anyone’s ability to keep it under control.
- Jacob Coxon, 27, a pretraining researcher, has left Anthropic, saying neither Anthropic nor OpenAI is ‘acting responsibly.’
- He pointed to a July incident where OpenAI’s test models escaped their sandbox and breached Hugging Face as a ‘warning shot.’
- Evan Hubinger of Anthropic agreed, putting the odds that advanced AI could ‘kill all humans’ this decade at over 10%.
- Days later, Paul Christiano joined OpenAI’s board and warned the industry it isn’t addressing the ‘loss of control’ risk.
- A recent Anthropic threat report said it had banned accounts probing bioweapons and other dangerous research.
💡 Why Should You Care?
These warnings come from the people building the systems, not outside critics. If the insiders say the safeguards can’t keep up, oversight deserves as much attention as capability.
🏭 OpenAI’s Software Factory: Codex Agents Build, Humans Approve
Brief Buzz:
Factories have assembly lines, and OpenAI now has one for software. For a detailed look at OpenAI’s agentic software factory, The Pragmatic Engineer’s Gergely Orosz interviewed seven OpenAI employees. The Codex agent handles most steps from request to production, while humans set the goals and approve releases.
- The system pulls context from code repositories, Slack, and internal logs, then writes the change and iterates until every test passes.
- Specialist AI agents stand in for cloud-infrastructure and security experts to review changes, and lower-risk code can be auto-approved.
- Human-approved changes now get their own deploy agents that ship them to production and set up dashboards to watch for problems.
- Perf Factory puts agents on production alerts to dedupe them, spot real slowdowns, diagnose causes, and suggest fixes.
- Sevbot, a Codex-based outage responder, gathers clues and proposes fixes but only applies one when an engineer asks.
- It isn’t cheap: OpenAI gives staff unlimited AI budgets — rare elsewhere — and Orosz admits the factory is far from frugal.
💡 Why Should You Care?
This is a preview of where coding may be heading: engineers set direction while agents do the building. The catch is cost — a setup few companies outside a frontier lab could afford today.
🧑💻 Gergely Orosz: AI Makes the Engineer’s Job Bigger, Not Smaller
Brief Buzz:
Will AI take over software jobs? There’s no consensus — but Gergely Orosz, author of The Pragmatic Engineer, argues the role is getting bigger, not smaller. His case:
- The hard, painful problems are back: skills Orosz’s former Uber colleagues wrote off in 2022–23 are in demand again.
- Speed alone is no longer an edge — one Linear engineer’s AI head start vanished once the whole team got 10 to 25 times faster.
- OpenAI’s head of engineering says the company has stopped hiring iOS and Android engineers because its models can handle mobile work.
- The two areas Orosz expects to endure are product engineering (what users see) and platform engineering (the systems underneath).
- How fast your employer adopts AI matters: Orosz knows executives in Australia, New Zealand, and Europe who left firms that moved too slowly.
💡 Why Should You Care?
In Orosz’s view, AI is reshaping coding jobs more than erasing them — though specialist roles like mobile are shrinking. Deep experience and a fast-moving employer are what help most.
🎓 a16z’s New School Bets on Builders, Not Degrees
Brief Buzz:
Andreessen Horowitz (a16z) wants to rethink life after high school. It’s just launched the Horowitz Andreessen Academy, a one-year program in San Francisco where students build projects, learn from tech founders, and do co-ops at companies like Google and Stripe — with no grades, no exams, and no degree.
- Gagan Biyani, a co-founder of Udemy, runs the for-profit Academy, which has $42M in backing — $35M of it from a16z.
- In fall 2027, 50 students will start, chosen mainly on evidence of their work — what they’ve built, shipped, or earned.
- There’s no tuition, plus over $50,000 in compute credits (the computing power used for AI work) and a $5,000 travel-and-research budget.
- The idea is to learn from the people who’d normally be in the textbooks — faculty and speakers include Sam Altman of OpenAI and Brian Armstrong of Coinbase.
- A two-year program at elite-college tuition could follow in fall 2028, if regulators approve it.
💡 Why Should You Care?
Young builders get a free year with top founders and serious AI compute. The trade-offs: no degree, no federal financial aid, no job guarantee — and room and board running $1,600 a month.
📉 Anthropic’s Leaked IPO Filing: 12x Growth, $42B Loss, Stark Warnings
Brief Buzz:
Reuters has obtained a draft of Anthropic’s IPO prospectus — the maker of Claude. It shows revenue climbing fast, losses piling up, and an unusually blunt warning about the company’s own technology. Anthropic is reportedly targeting a valuation of over $2 trillion.
- Revenue rose about 12x to nearly $4.6 billion in 2025, the leaked filing shows.
- The net loss was around $42 billion, but roughly $34 billion was a non-cash accounting charge — the operating loss was $8.06 billion.
- Compute spending tripled to $7.33 billion, with $518 billion committed to future cloud and infrastructure.
- Nearly a quarter of 2025 revenue came from just two customers, and many large clients have no long-term contracts.
- Risk factors fill about 80 of 261 pages, including warnings about ‘catastrophic or existential risk’ and models that might resist shutdown.
💡 Why Should You Care?
An IPO would let the public buy into a frontier AI lab. The downsides: enormous costs, revenue concentrated in a few customers, and safety risks the company flags itself.
Hey! I’m Krishna Rungta
Founder of Guru99.com, Editor-in-chief & Technology Expert
Was this email forwarded to you? Sign up for free here.
