In partnership with

A new version of OpenAI’s Codex is powered by a new dedicated chip

Key Takeaway: OpenAI unveiled a research preview of GPT-5.3-Codex-Spark, a real-time “conversational” coding model that’s reportedly ~15× faster than GPT-5.3-Codex—at the cost of lower benchmark performance and weaker cybersecurity capability.

More Insights:

• Built for rapid, interruptible iteration (quick edits, fast back-and-forth), not long-running autonomous coding “agents.”

• Major latency cuts: ~80% lower roundtrip overhead, ~50% faster time-to-first-token, plus a persistent WebSocket connection.

• Runs on Cerebras WSE-3 wafer-scale chips to prioritize speed and responsiveness.

• Comes with usage constraints: initially limited to Pro ($200/mo) users, and may slow/queue during high demand.

• “Fast, but less capable”: underperforms the full model on agentic software-engineering benchmarks and is explicitly not near OpenAI’s “high capability” cybersecurity threshold.

Why it matters: Speed changes how developers work—but when “good enough” code arrives 15× faster, the real risk isn’t just more bugs, it’s faster confidence in the wrong answer, which can quietly turn velocity into fragility.

Dictate prompts and tag files automatically

Stop typing reproductions and start vibing code. Wispr Flow captures your spoken debugging flow and turns it into structured bug reports, acceptance tests, and PR descriptions. Say a file name or variable out loud and Flow preserves it exactly, tags the correct file, and keeps inline code readable. Use voice to create Cursor and Warp prompts, call out a variable like user_id, and get copy you can paste straight into an issue or PR. The result is faster triage and fewer context gaps between engineers and QA. Learn how developers use voice-first workflows in our Vibe Coding article at wisprflow.ai. Try Wispr Flow for engineers.

Start flowing free

IBM will hire your entry-level talent in the age of AI

Key Takeaway: IBM says it will triple entry-level hiring in the US in 2026 by redesigning junior roles so new grads add value in an AI-heavy workplace instead of competing with automation.

More Insights:

• Hiring expansion is described as “across the board,” spanning many departments, not just tech.

• IBM rewrote entry-level job descriptions because AI can now handle much of what junior roles did 2–3 years ago.

• Junior software developers are shifting from routine coding to customer-facing work and higher-context problem solving.

• Entry-level HR staff increasingly step in when chatbots fail, correcting outputs and escalating to managers when needed.

• IBM argues cutting junior hiring now could create a future mid-level leadership gap, forcing expensive poaching later; Dropbox echoed the theme, boosting early-career programs 25% to leverage younger workers’ AI fluency.

Why it matters: AI isn’t just “removing jobs”—it’s rewriting the career ladder, and IBM’s move is a bet that companies will either train adaptable talent now or pay a premium later for experience they failed to cultivate.

Spotify says its best developers haven’t written a line of code since December, thanks to AI

Key Takeaway: Spotify says its best engineers haven’t written code since December because an internal AI system (“Honk,” powered by Claude Code) can generate fixes and features, deploy builds, and let humans review/merge—dramatically accelerating product velocity.

More Insights:

• Spotify claims “Honk” enables remote, real-time deployment via generative AI workflows (even from a phone during a commute).

• Co-CEO Gustav Söderström described a loop where an engineer requests a fix/feature in Slack, gets a new build back, then merges to production.

• Spotify highlighted shipping 50+ app features/changes in 2025, with newer AI-driven launches like Prompted Playlists, Page Match for audiobooks, and About This Song.

• The company argues its music-related dataset is uniquely defensible because many music questions are preference-based, not factual like Wikipedia.

• On AI-generated music, Spotify says it supports metadata labeling for how tracks are made while continuing to police spam.

Why it matters: If Spotify’s workflow is real and scalable, the “tipping point” isn’t AI writing code—it’s AI collapsing the entire build-to-deploy cycle into a supervised, near-instant feedback loop, shifting the engineer’s core job from typing to directing, validating, and owning outcomes.

Other stuff

• OpenAI’s President Gave Millions to Trump. He Says It’s for Humanity

• Rules of mysterious ancient Roman board game decoded by AI

• How private equity’s big bet on software was derailed by AI

• OpenAI’s Biggest Challenge Is Turning Its A.I. Into a Cash Machine

• Seedance 2 pulled as it unexpectedly reconstructs voices accurately from face photos.

• Brace For AI Or Leave: Google Tells Employees

• Anthropic to cover costs of electricity price increases from its data centers

• OpenAI Executive Who Opposed ‘Adult Mode’ Fired for Sexual Discrimination

Hope you enjoyed today's newsletter

⚡️ Join over 300,000 people using the Superpower ChatGPT extension on Chrome and Firefox.