Hey friend,

Owain here. Today, we’ll explore

• My personal framework for writing prompts

• OpenAI Codex: An new AI agent for software engineers

• Notion’s latest AI upgrades

• Windsurf's new AI models

Let’s dive in.

Tutorial: The 10-Part Perfect Prompt Template

If you struggle with how to structure your prompts this should help. I use the following structure to think clearly about how to design effective prompts for AI.

While I list 10 sections, not all of them are necessary for every task. The framework here helps you think clearly. Once you have outlined your prompt you can pass it to AI to refine and improve your prompt further.

1. GOAL / OBJECTIVE

State your ultimate aim. What are you trying to accomplish? Example: Create an engaging Twitter thread from a technical article.

2. PERSONA

Who should the AI be? This shapes everything. Example: Act as a talented copywriter who excels at breaking down complex topics.

3. TASK

The specific action(s) you want. Use clear verbs. Example: Analyse the article. Identify 5-7 key takeaways. Draft tweets for each.

4. TARGET AUDIENCE

Who is the final output for? Example: Tech enthusiasts curious about AI but not experts.

5. CONTEXT

Provide all necessary background information. Example: Here's the article text: [Insert between tags like <article>...</article>]

6. KEY SPECIFICATIONS

Detailed guidelines or steps to follow. Example: 1. Read the article. 2. Identify three main arguments. 3. For each, provide a supporting quote.

7. EXAMPLES

Show, don't just tell. Provide samples of desired output.

8. TONE / STYLE

The writing style or attitude for the response. Example: Friendly, enthusiastic, and conversational. Use emojis sparingly.

9. OUTPUT FORMAT

How should the response be structured? Example: Markdown with H2 headings and bullet points for sub-items

10. RULES & CONSTRAINTS

Final limitations and requirements. Example: Keep under 500 words. ALWAYS include a call-to-action at the end.

Pro Tips:

• Start simple: Begin with Goal, Task, and Context for basic prompts.

• Be specific: The more detail you provide, the better your results.

• Iterate: Your first prompt is rarely perfect. Refine based on results.

• Use delimiters: Separate sections with ### or XML-style tags.

• Meta-prompt: Ask the AI itself how to improve your prompt!

Remember: Effective prompting is a skill. Practice this framework, and you'll soon be getting AI outputs that exceed your expectations.

AI News

Here’s my pick of AI news this week:

1. OpenAI's Codex

OpenAI has launched Codex, a cloud-based software engineering agent powered by codex-1, enabling developers to simultaneously tackle multiple coding tasks like bug fixes, feature development, and codebase analysis in isolated cloud environments.

• Codex works in sandboxed cloud VMs supporting parallel tasks that take 1-30 minutes each, providing transparent logs and test results while iteratively improving code quality.

• Early enterprise adopters including Cisco and Temporal report faster shipping times and improved code quality through debugging, refactoring, and feature development capabilities.

• Currently available to ChatGPT Pro, Enterprise, and Team subscribers, OpenAI plans to implement rate limits and credit purchases for broader access, while developing Codex into a more capable "virtual teammate."

2. Notion AI Suite Slashes Enterprise Costs By 60%

Notion has launched AI for Work, a suite of enterprise AI tools integrated into its workspace, featuring AI Meeting Notes, Enterprise Search, and Research Mode. Business and Enterprise plans now include unlimited Notion AI with cross-platform search capabilities, reducing AI costs by 40-60% compared to standalone tools.

• AI Meeting Notes automatically transcribes and summarises meetings, while Enterprise Search connects with Slack, Teams, and Google Drive, breaking down information silos across applications.

• The suite provides direct access to GPT-4.1 and Claude 3.7 within Notion, intelligently switching between models based on specific task requirements.

• Priced at $30/user/month as part of the Business plan, Notion AI for Work undercuts comparable AI tool stacks by over $420 per user annually, with early adopters including OpenAI, Ramp, and Vercel.

3. Windsurf's New AI Models Match Claude While Cutting Costs By 40%

Windsurf has launched SWE-1, its first family of proprietary AI models designed specifically for the entire software engineering lifecycle, featuring a three-tier approach and "flow awareness" technology that enables contextual collaboration across development environments.

• SWE-1 introduces "flow awareness" that creates a shared timeline between users and AI, allowing seamless collaboration across editors, terminals, and browsers throughout the development process.

• Internal benchmarks position the flagship SWE-1 model competitively with Claude 3.5 Sonnet while being 40% cheaper to operate, outperforming all non-frontier and open weight models.

• The family includes three models: the full-size SWE-1 for paid users, SWE-1-lite replacing Cascade Base for all users, and SWE-1-mini for passive code prediction tasks.

Your Opinion Matters

What did you think of today’s email? Your feedback helps me create better emails for you!

Got more feedback or want me to cover a specific topic? Reply to this email and let me know.

Owain