Frederic Bazin

Entrepreneurship, Software Engineering, Product development, Community Building.

[caffeine @caffeineai]

If you’re in 🌴 Bali, join us for a chance to experience the Self-Writing Internet with Caffeine, the first complete tech stack designed for AI, where humans build through conversation only.

🇮🇩 Bali
🗓️ August 23rd, 2025
⏱️ 09:30 – 14:00

🎟️ Register: 👇
https://t.co/qmwSRt7SwH

[Frederic Bazin @bazinfrederic]

@samlafer What is missing? GitHub PR do support squash merge

[Frederic Bazin @bazinfrederic]

Nice upgrade to open telemetry with profiling. https://t.co/vpASVXuS0B

[Andrew Ng @AndrewYNg]

I think AI agentic workflows will drive massive AI progress this year — perhaps even more than the next generation of foundation models. This is an important trend, and I urge everyone who works in AI to pay attention to it.

Today, we mostly use LLMs in zero-shot mode, prompting a model to generate final output token by token without revising its work. This is akin to asking someone to compose an essay from start to finish, typing straight through with no backspacing allowed, and expecting a high-quality result. Despite the difficulty, LLMs do amazingly well at this task!

With an agentic workflow, however, we can ask the LLM to iterate over a document many times. For example, it might take a sequence of steps such as:
- Plan an outline.
- Decide what, if any, web searches are needed to gather more information.
- Write a first draft.
- Read over the first draft to spot unjustified arguments or extraneous information.
- Revise the draft taking into account any weaknesses spotted.
- And so on.

This iterative process is critical for most human writers to write good text. With AI, such an iterative workflow yields much better results than writing in a single pass.

Devin’s splashy demo recently received a lot of social media buzz. My team has been closely following the evolution of AI that writes code. We analyzed results from a number of research teams, focusing on an algorithm’s ability to do well on the widely used HumanEval coding benchmark. You can see our findings in the diagram below.

GPT-3.5 (zero shot) was 48.1% correct. GPT-4 (zero shot) does better at 67.0%. However, the improvement from GPT-3.5 to GPT-4 is dwarfed by incorporating an iterative agent workflow. Indeed, wrapped in an agent loop, GPT-3.5 achieves up to 95.1%.

Open source agent tools and the academic literature on agents are proliferating, making this an exciting time but also a confusing one. To help put this work into perspective, I’d like to share a framework for categorizing design patterns for building agents. My team AI Fund is successfully using these patterns in many applications, and I hope you find them useful.

- Reflection: The LLM examines its own work to come up with ways to improve it.
- Tool use: The LLM is given tools such as web search, code execution, or any other function to help it gather information, take action, or process data.
- Planning: The LLM comes up with, and executes, a multistep plan to achieve a goal (for example, writing an outline for an essay, then doing online research, then writing a draft, and so on).
- Multi-agent collaboration: More than one AI agent work together, splitting up tasks and discussing and debating ideas, to come up with better solutions than a single agent would.

I’ll elaborate on these design patterns and offer suggested readings for each next week.

[Original text: https://t.co/y4McIAjD2m]

[Frederic Bazin @bazinfrederic]

@NotionHQ @NotionCalendar Love the notion calendar. Been a huge fan of cronapp.

But when you share an availability link on whatsapp....
The social thumbnail and caption shows ad :(

I hope we can see something more useful soon.

In 2009, the email signature "send from my IPhone" was not so embarassing.

[Frederic Bazin @bazinfrederic]

@NotionHQ @NotionCalendar Love the notion calendar. Been a huge fan of cronapp.

But when you share an availability link on whatsapp....
The social thumbnail and caption shows ad :(

I hope we can see something more useful soon.

In 2009, the email signature "send from my IPhone" was not so embarassing.

[🦋 AnneInContext @anneincontext]

How Life Of Indie Hackers Like?

🎙️ Our honored guests are Evan You (@evanyou) and Erwin (@Erwin_AI ) this time.

💻 Evan, dedicating full-time hours to @vuejs and @vite_js , is also an indie hacker. Ever wondered what his daily life looks like?

🛠️ Erwin, a former backend developer turned indie hacker, built his journey in public, @tailscan and @BootstrFM. Dive into his progress and experiences!

📢 Have burning questions for Evan and Erwin? Submit them now via https://t.co/ZSVbS9tjDx, and they might spill the beans during our live session!

Set a reminder by clicking https://t.co/UQVP3d625P.

#indiehack #opensource #TechNews #TechTips

[Frederic Bazin @bazinfrederic]

@frantzfries Any app can replicate more or less document editing features.
But navigation and performance is another challenge that would require rebuilt from the ground up

[Frederic Bazin @bazinfrederic]

The island of gods is on fire!

#waste #landfill #sustainability

https://t.co/XOF1lczXF8

[Frederic Bazin @bazinfrederic]

4x performance increase over node.js
https://t.co/E2z4xPj9v4

[Frederic Bazin @bazinfrederic]

Up to 68,000x performance increase over standard Python
https://t.co/vg41ngRpeK

[Frederic Bazin @bazinfrederic]

@petervanderdoes @raymondh @pycharm Looks very thorough

[Frederic Bazin @bazinfrederic]

You remember this time when you hated your phone because the phone manufacturer was designing the software.
Well , today the liberation is happening to your car

[Martin Fowler @martinfowler]

NEW POST

My colleague Sarah Taraporewalla begins an article to describe an approach to building technology strategy. It identifies common strategic directions, using them to indicate the investigations needed to understand the technology implications

https://t.co/7DHideJCl0

[Frederic Bazin @bazinfrederic]

nice visual on red flags for #productmanagement

[Paweł Huryn @PawelHuryn]

Be careful. Most "products" are, in fact, projects.

9 red flags (and how it should work):

1. Large PRD: You start an initiative by documenting everything.

2. Feature factory: Implement the requirements. Don't ask why.

3. Waterfall: All the requirements are collected in the "initial phase."

4. Gatt roadmap: A time-based, feature-based roadmap.

5. No discovery: No need to validate ideas before implementing them.

6. No designer: There is no Product Designer on the team.

7. No analytics: You have no idea how people use your product.

8. Customer in charge: Powerful customer(s) make all the decisions.

9. No strategy: You try to maximize sales by satisfying all customers and grasping every opportunity.

-

Here is a better way:

1. Your cross-functional team is empowered to solve the problems.

2. PM, Product Designer, and Lead Engineer perform Product Discovery together. Continuously.

3. You have an outcome-based roadmap. Preferably Now-Next-Later.

4. If you commit to a date, you do it rarely and only after the Discovery. You never commit too early.

5. You manage the value, usability, feasibility, and viability risks by experimenting.

6. The riskiest assumptions are tested before the implementation.

7. Choosing, instrumenting, and tracking the right metrics is key.

8. You ship incrementally, measure the outcomes and learn from it.

9. Tradeoffs are essential. What you do, but also what you don't. You respect your market and the unique value proposition.

-

And if your product hasn't been launched yet:

1. Discover the market and define a unique value proposition, business model, initial vision, and strategy.

2. Test your business idea with the help of MVP prototypes. Before the implementation.

3. You define the go-to-market strategy and validate key assumptions. Messaging included.

4. You can't rely on product analytics before launching the product, so you rely more on customer interviews and data from your experiments.

5. The Product Trio performs the Initial Product Discovery, like in an existing product. You always need a Product Designer and Lead Engineer.

6. Once you ship, use product analytics and apply Continuous Product Discovery.

-

Hope that helps.

What are your thoughts?

-

P.S. It's just 1 of 6 free issues I published today in my newsletter. The link is under my profile: @PawelHuryn

[🦋 AnneInContext @anneincontext]

Happening in 24 hours ☺️

[Martin Fowler @martinfowler]

NEW POST

While coding assistants like Copilot can improve speed and flow, they can also disrupt it. Birgitta Böckeler looks at two ways in which they can get in the way: amplifying bad or outdated practices, and review fatigue.

https://t.co/UerXbksAqh?

[Frederic Bazin @bazinfrederic]

if data scientists can now generate clean code, software engineers will soon be out of job.

[Lior Alexander @LiorOnAI]

Great news. Jupyter just added a LLM-based chatbot to their environment.

The new AI bot allows you to:
▸ Generate entire notebooks from a prompt
▸ Generate code
▸ Fix errors
▸ Summarize content
▸ Ask questions about local files

Jupyter AI connects Jupyter with large language models (LLM) from providers such as AI21, Anthropic, AWS, Cohere, and OpenAI

pip install jupyter-ai

[Frederic Bazin @bazinfrederic]

And I thought the WhatsApp roadmap had been accelerating (or at least waking up) since November. This is no comparison with this. Well done @x team!

[X @X]

The vision for X has been unfolding over the last 9 months – and there’s so much more on the horizon. Today we’re taking a moment to celebrate the progress and thank our teams and all our partners. https://t.co/azXPbe8IpO

[Andy Zou @andyzou_jiaming]

🚨We found adversarial suffixes that completely circumvent the alignment of open source LLMs. More concerningly, the same prompts transfer to ChatGPT, Claude, Bard, and LLaMA-2…🧵

Website: https://t.co/ja2FPw9aad
Paper: https://t.co/1q4fzjJSyZ