Think Your Job Is Safe From AI? Think Again
Cheaper AI is getting smarter, OpenAI is advancing scientific discovery, and Google is upgrading AI image generation
In this SunBrief
Your Best Closers Shouldn’t Be Stuck Doing $20 Tasks
Your top salespeople should be spending their time creating revenue. Instead, they’re updating CRM records, scheduling meetings, chasing signatures, and managing the administrative work surrounding every deal. Those tasks matter, but they’re not the highest-value use of your most expensive talent. The free guide _5 Reasons You Aren’t Hitting Quota: The Sales Operations Gap Examined _helps sales leaders identify the operational gaps that can pull reps away from selling. BELAY supports sales teams by matching them with U.S.-based Sales Assistants who manage the work around the sale, from CRM management and lead routing to follow-up coordination.
Download the free guide 5 Reasons You Aren’t Hitting Quota01 / Jobs watch
These U.S. states have the most jobs at risk from AI
A new analysis maps where American workers face the most AI disruption. What interests me is how heavily the exposure falls on skilled, white-collar jobs.
- Washington leads at 5.7% of workers in highly AI-exposed roles, followed by Virginia (4.6%) and Washington, D.C. (4.5%).
- California has the most exposed jobs, with 724,000 positions, followed by Texas at 500,000.
- Math, programming and writing jobs face significant exposure, but that could mean changing responsibilities rather than disappearing positions.
Why it matters: I'd watch hiring and wages, not just layoffs. AI could change which skills employers pay for long before entire jobs disappear.
Read more ↗What is the best response to rising AI exposure at work?
Market Snapshot
| Company | Price ($) | 5D (%) | YTD (%) | ||
|---|---|---|---|---|---|
|
|
347.37 | -0.11 | 10.16 | ||
|
336.67 | 2.02 | 24.23 | ||
|
721.31 | -0.99 | 10.90 | ||
|
529.76 | 1.90 | 12.01 | ||
|
259.92 | 3.30 | 14.75 | ||
|
237.47 | 3.26 | 25.75 |
04 / Model watch
Claude’s cheapest model just got scary good
Anthropic launched Claude Haiku 5.5, its fastest and cheapest small model yet. The bigger story is how capable low-cost AI is becoming.
- It starts at $0.10/$0.50 per million input/output tokens for prompts under 100K, around 90% cheaper than Haiku 4.5.
- Anthropic’s tests put it ahead of GPT-6 Luna on knowledge work, computer use and its main FrontierCode coding benchmark.
- Haiku 5.5 is available now for summaries, browser tasks, customer support and subagents.
Why it matters: Haiku makes repetitive AI work much cheaper, reducing the need to use expensive frontier models for every task.
Read more ↗Are frontier models becoming unnecessary for most everyday AI workloads?
05 / Research watch
OpenAI’s unreleased model wrote 722 math papers
OpenAI published 722 mathematical manuscripts generated by an unreleased frontier model. The number is impressive, but I’m more interested in whether mathematicians decide the work is actually new.
- The papers cover 372 families of results, with OpenAI estimating roughly three hours of ChatGPT Pro level thinking per result.
- Many proofs were checked with Lean, giving researchers a way to verify the logic. Others still need independent review.
- Mathematicians want more transparency, including prompts and reasoning, and have warned labs against turning research discoveries into model marketing.
Why it matters: Producing hundreds of proofs is one thing. Proving the AI contributed genuinely new ideas, rather than remixing human work, is the harder test.
Read more ↗What would be the bigger breakthrough?
05 / AI highlights
AI in Brief
PRODUCTIVITY
Claude Can Now Edit Your Google Docs for You
Claude can now work inside Docs, Sheets and Slides, making edits, formulas, charts and presentations without switching apps.
IMAGES
Google Just Gave Nano Banana a Big Upgrade
Nano Banana 2.1 improves text, character consistency, precise editing and world knowledge for more polished AI-generated images.
AGENTS
Meta Wants AI Agents to Log In Like You Do
Meta and Sierra, backed by Walmart and Stripe, want a shared login standard for AI agents after Amazon blocked Muse.
OPEN WEIGHTS
Mistral's best AI model yet takes on DeepSeek and Qwen
Mistral Large 4 is a 1-trillion-parameter multimodal model built for coding, agents and enterprise work, with open weights coming this month.
On Our Radar
- OpenAI is bringing visual ads to ChatGPT image generation
- Google opens SynthID tool to everyone to spot AI-generated media
- OpenAI will watermark ChatGPT and Codex text in the EU
- Reflection unveils Beam, a 501B open-weight AI model
- New AI assistant Tab emerges from stealth at $300M valuation
- AI optimism beats concern in 34 of 37 countries
Last Week’s Poll Result
What matters most when choosing between Gemini, ChatGPT, and Claude?
| Best overall intelligence | 55.88% |
| Lowest cost | 14.71% |
| Strongest agent performance | 20.59% |
| Best ecosystem and tools | 8.82% |
Where could AI have the biggest impact on historical research?
| Decoding old ciphers | 28.57% |
| Reading damaged manuscripts | 28.57% |
| Translating forgotten languages | 28.57% |
| Searching huge archives | 14.29% |
Which DevDay update feels most important?
| Always-on Dots agents | 28.57% |
| GPT-6.1 Sol pricing | 42.86% |
| Shared ChatGPT Space | 14.29% |
| App integrations | 14.29% |
FROM SMARTER WITH AI
Practical guides, walkthroughs and adaptable prompts to help you put AI to work. $29 once. No recurring subscription.
Explore the AI Library →We’d love to hear from you!
Thanks for reading. See you next Sunday.
