- Horizon AI
- Posts
- Use This Prompt to Let Codex Turn Your Repetitive Work into Automations and Skills🚀
Use This Prompt to Let Codex Turn Your Repetitive Work into Automations and Skills🚀
OpenAI Offers $445000 Salary for AI Safety Research Role 💰
Gina Acosta
May 25, 2026
In partnership with
Welcome to another edition of Horizon AI,
Repetitive work ends up building quietly across sessions. In today's issue, we'll share a Codex prompt that scans your recent work, Memories, and Chronicle to identify patterns, reuse what already exists, and create useful skills, subagents, or automations.
Let’s jump into it!
Read Time: 4.5 min
Here's what's new today in the Horizon AI
OpenAI Is Paying Up to $445,000 for an AI Safety Role
Google DeepMind Cracks 9 of 353 Open Erdős Math Problems
AI Tutorial: Let Codex Turn Your Repetitive Work into Automations and Skills
AI Tools to check out
AI Findings/Resources
The Latest in AI and Tech 💡
AI News
OPENAI
OpenAI Is Paying Up to $445,000 for an AI Safety Role
OpenAI is hiring new safety researchers as the company prepares for a future where AI systems may be capable of training and improving newer versions of themselves.
Details:
OpenAI recently posted a research role offering compensation between $295,000 and $445,000 focused on preparing for “recursive self-improvement,” where AI systems could potentially automate parts of their own development.
The company says the role involves anticipating risks that may not exist yet, which is why it is looking for researchers who are both technically strong and “tasteful and strategic.”
OpenAI CEO Sam Altman has previously said the company hopes to eventually build a fully automated AI researcher capable of contributing to AI development itself.
The new safety role could involve protecting models from threats like data poisoning, monitoring progress toward automating technical work, and studying how advanced models reason internally.
Researchers and AI leaders increasingly believe self-improving AI systems may become possible within the next several years, though opinions differ sharply on the risks and timelines involved.
TOGETHER WITH WISPR FLOW
Your prompts are leaving out 80% of what you're thinking.
When you type a prompt, you summarize. When you speak one, you explain. Wispr Flow captures your full reasoning — constraints, edge cases, examples, tone — and turns it into clean, structured text you paste into ChatGPT, Claude, or any AI tool. The difference shows up immediately. More context in, fewer follow-ups out.
89% of messages sent with zero edits. Used by teams at OpenAI, Vercel, and Clay. Try Wispr Flow free — works on Mac, Windows, and iPhone.
Google DeepMind Cracks 9 of 353 Open Erdős Math Problems
Google DeepMind has unveiled AlphaProof Nexus, a new AI framework that autonomously solved several long-standing mathematical problems combining LLM-driven proof generation with machine verification.
Details:
The system solved 9 out of 353 open Erdős problems, including two questions that had remained unsolved for more than 50 years.
AlphaProof Nexus also proved 44 open conjectures from the Online Encyclopedia of Integer Sequences, settled a long-standing algebraic geometry problem, and improved known bounds in convex optimization.
The framework uses Gemini 3.1 Pro to generate mathematical proof steps in Lean, a formal programming language designed for machine-verifiable mathematics.
Instead of relying purely on natural language reasoning, the system continuously checks each proof step through the Lean compiler, using compiler feedback to guide future attempts and reduce logical errors.
According to the research paper, inference costs ran just a few hundred dollars per problem.
The news comes just days after OpenAI claimed its new reasoning model produced an original mathematical proof disproving an 80-year-old Erdős conjecture.
AI Tutorial
Let Codex Turn Your Repetitive Work into Automations and Skills
Paste this prompt into Codex:
"Look back over my recent work from the last 30 days, or all available history if shorter, and identify repeated manual workflows worth packaging.
Use available evidence in this order:
- Recent Codex sessions and task summaries.
- Codex Memories and rollout summaries to find patterns repeated across sessions.
- Chronicle, if enabled, to spot repeated work outside Codex. Use Chronicle for discovery only; confirm important details in the relevant source system when possible.
- Existing skills, custom agents, and automations, so you reuse or extend what already exists instead of duplicating it.
Look broadly for work that is repeated, time-consuming, error-prone, context-heavy, or benefits from a consistent process. Include workflows across coding, research, writing, planning, communication, operations, analysis, and personal administration.
Only act on a candidate when it:
- occurred at least twice, or is clearly likely to recur and costly to repeat;
- has stable inputs, a repeatable procedure, and a clear output or stopping condition;
- would materially improve speed, quality, consistency, or reliability;
- is not already adequately covered.
Choose the smallest appropriate form:
- Skill: a reusable workflow or playbook.
- Custom subagent: a bounded specialist role or investigation task suitable for delegation.
- Automation: a scheduled or recurring check, report, reminder, or monitor.
- Skip: work that is too one-off, ambiguous, sensitive, or poorly evidenced to package.
First produce a compact shortlist with:
- repeated workflow
- supporting evidence and dates
- frequency/confidence
- recommended form: skill, subagent, automation, extend existing, or skip
- why it is or is not worth creating
Then create only the high-confidence missing items. Keep them narrow, practical, source-aware, and easy to validate. Do not create speculative, overlapping, or overly broad assets.
Finish with:
- what you created or extended
- what you deliberately skipped
- what needs more evidence before packaging"Source: @reach_vb on X
AI Tools to check out
🔥 Wizstar: A one-stop AI video creation platform built for creators and e-commerce brands. Just drop in a product link or a single sentence and the AI agent handles everything else.
🚀 Polsia: An autonomous AI system that plans, codes, and markets your company 24/7.
📞 Shadow: A real-time AI wingman for high-stakes calls that helps you ask better questions, never miss key details, and turn every conversation into clear next steps.
✅ GetThis: Generate tasks from voice, text, or screenshots.
🧠 Kanwas: An open-source brain for your team. It gives teams & agents one place to create, edit, share and compound product context.
TOGETHER WITH WISPR FLOW
Stop re-prompting. Say it right the first time.
Voice-first prompts preserve the nuance you cut when typing. Speak once, paste into any AI tool, get results that don't need a follow-up. 89% of messages sent with zero edits.
AI Findings/Resources
👩🏫 Teachers are becoming more AI-savvy, embedding hidden instructions in assignments to detect AI-generated responses
🤔 Microsoft canceled its Claude Code licenses because the bill got too high, and they may not be the last company to do the math
👀 Pope XIV says the church and Anthropic, will work together to "find the way for humanity, in this time of artificial intelligence."
✍ You can use ChatGPT to fill out forms easier by just telling it what to write
The latest in AI and Tech
Anthropic has reportedly told investors it expects to more than double quarterly revenue to around $10.9 billion while reaching an operating profit for the first time. The milestone could strengthen the company’s position against rival OpenAI, though profitability may not last through the year due to rising compute costs.
A startup claims to have developed an $118 AI collar that translates barks and meows into complete sentences. The company has gone viral on X, but there is a lot of skepticism around the validity of its claims. That hasn't stopped 10,000 people from pre-ordering it.
The White House has approved a request for $9 billion to help U.S. intelligence agencies expand their AI computing capabilities, though the funding still requires congressional approval. According to reports, agencies like the CIA and the NSA currently lack enough computing power to run the latest AI models effectively.
The company is reportedly in early talks to use AI servers powered by Microsoft’s Maia 200 chips as demand for Claude continues to grow.
Federal prosecutors have charged two men accused of posting thousands of nonconsensual AI-generated intimate deepfakes online under the Take It Down Act.
That’s a wrap!
We'd love to hear your thoughts on today's email!Your feedback helps us improve our content |
Not subscribed yet? Sign up here and send it to a colleague or friend!
See you in our next edition!
Gina 👩🏻💻