Jonathan
@jonathanlow.bsky.social
Juggling parenthood and a startup - Mistle (mistle.dev)
Been landing up in contextual limbo when I multi-task too aggressively. Managing context requires a lot of intention to keep up.
October 29, 2025 at 7:16 AM
Been landing up in contextual limbo when I multi-task too aggressively. Managing context requires a lot of intention to keep up.
I’ve been using GPT-5 Codex to critique its own outputs with a recursive feedback loop.
With mistle.dev side threads, I run critique → reflection → revision cycles.
It actually works pretty well.
With mistle.dev side threads, I run critique → reflection → revision cycles.
It actually works pretty well.
October 28, 2025 at 6:38 AM
I’ve been using GPT-5 Codex to critique its own outputs with a recursive feedback loop.
With mistle.dev side threads, I run critique → reflection → revision cycles.
It actually works pretty well.
With mistle.dev side threads, I run critique → reflection → revision cycles.
It actually works pretty well.
Vibed out a Chrome Extension that helps to write X replies in less than an hour.
The vibing is the easy part.
The hard part is writing the prompts to generate responses that reflects my style and thinking.
The vibing is the easy part.
The hard part is writing the prompts to generate responses that reflects my style and thinking.
October 27, 2025 at 9:23 AM
Vibed out a Chrome Extension that helps to write X replies in less than an hour.
The vibing is the easy part.
The hard part is writing the prompts to generate responses that reflects my style and thinking.
The vibing is the easy part.
The hard part is writing the prompts to generate responses that reflects my style and thinking.
Relaunching Zapi as mistle.dev
Mistle is Cursor for non-technical builders.
We built Mistle to help non-technical builders learn, understand, and get better as they build.
We've got a few early users who jumped from Cursor to Mistle.
If you're non-technical, give Mistle a try.
Mistle is Cursor for non-technical builders.
We built Mistle to help non-technical builders learn, understand, and get better as they build.
We've got a few early users who jumped from Cursor to Mistle.
If you're non-technical, give Mistle a try.
October 24, 2025 at 5:46 PM
Relaunching Zapi as mistle.dev
Mistle is Cursor for non-technical builders.
We built Mistle to help non-technical builders learn, understand, and get better as they build.
We've got a few early users who jumped from Cursor to Mistle.
If you're non-technical, give Mistle a try.
Mistle is Cursor for non-technical builders.
We built Mistle to help non-technical builders learn, understand, and get better as they build.
We've got a few early users who jumped from Cursor to Mistle.
If you're non-technical, give Mistle a try.
We're launching Zapi as an AI Copilot for Product Teams.
With Zapi - Write and turn PRDs into prototypes in minutes.
How it works:
1. Prompt your idea.
2. Iterate till you get a working PRD
3. Zapi builds a hi-fi prototype according to your specs
4. Publish and share with your team
With Zapi - Write and turn PRDs into prototypes in minutes.
How it works:
1. Prompt your idea.
2. Iterate till you get a working PRD
3. Zapi builds a hi-fi prototype according to your specs
4. Publish and share with your team
June 13, 2025 at 10:01 AM
We're launching Zapi as an AI Copilot for Product Teams.
With Zapi - Write and turn PRDs into prototypes in minutes.
How it works:
1. Prompt your idea.
2. Iterate till you get a working PRD
3. Zapi builds a hi-fi prototype according to your specs
4. Publish and share with your team
With Zapi - Write and turn PRDs into prototypes in minutes.
How it works:
1. Prompt your idea.
2. Iterate till you get a working PRD
3. Zapi builds a hi-fi prototype according to your specs
4. Publish and share with your team
Before building an app, it's important to determine an initial direction and requirement.
But writing a doc is boring and time-consuming. So we built Zapi to one-shot a comprehensive product requirements document (PRD).
But writing a doc is boring and time-consuming. So we built Zapi to one-shot a comprehensive product requirements document (PRD).
May 14, 2025 at 8:47 AM
Before building an app, it's important to determine an initial direction and requirement.
But writing a doc is boring and time-consuming. So we built Zapi to one-shot a comprehensive product requirements document (PRD).
But writing a doc is boring and time-consuming. So we built Zapi to one-shot a comprehensive product requirements document (PRD).
UX often breaks not because it was bad but because the product grew.
Zapi started with 1 view mode. As we added more, we used a dropdown.
It worked until it didn’t.
Now with 7 modes, we moved to a sidebar for clarity.
UX isn’t set and forget. Revisit what used to work.
Zapi started with 1 view mode. As we added more, we used a dropdown.
It worked until it didn’t.
Now with 7 modes, we moved to a sidebar for clarity.
UX isn’t set and forget. Revisit what used to work.
May 14, 2025 at 4:25 AM
UX often breaks not because it was bad but because the product grew.
Zapi started with 1 view mode. As we added more, we used a dropdown.
It worked until it didn’t.
Now with 7 modes, we moved to a sidebar for clarity.
UX isn’t set and forget. Revisit what used to work.
Zapi started with 1 view mode. As we added more, we used a dropdown.
It worked until it didn’t.
Now with 7 modes, we moved to a sidebar for clarity.
UX isn’t set and forget. Revisit what used to work.
We're looking for beta testers!
Zapi turns your ideas into a specifications doc and then follows the plan to build your idea.
As a beta tester, you get 100 free messages.
Join in 2 easy steps:
1. Join our discord: discord.com/invite/2RQ2g...
2. Post in #general channel with your Zapi user id
Zapi turns your ideas into a specifications doc and then follows the plan to build your idea.
As a beta tester, you get 100 free messages.
Join in 2 easy steps:
1. Join our discord: discord.com/invite/2RQ2g...
2. Post in #general channel with your Zapi user id
May 13, 2025 at 9:37 AM
We're looking for beta testers!
Zapi turns your ideas into a specifications doc and then follows the plan to build your idea.
As a beta tester, you get 100 free messages.
Join in 2 easy steps:
1. Join our discord: discord.com/invite/2RQ2g...
2. Post in #general channel with your Zapi user id
Zapi turns your ideas into a specifications doc and then follows the plan to build your idea.
As a beta tester, you get 100 free messages.
Join in 2 easy steps:
1. Join our discord: discord.com/invite/2RQ2g...
2. Post in #general channel with your Zapi user id
Tackling usability issues with navigation.
Which is more UX friendly?
Top version - collapsible sidebar menu
Bottom version - dropdown menu
#buildinpublic
Which is more UX friendly?
Top version - collapsible sidebar menu
Bottom version - dropdown menu
#buildinpublic
May 6, 2025 at 11:27 AM
Tackling usability issues with navigation.
Which is more UX friendly?
Top version - collapsible sidebar menu
Bottom version - dropdown menu
#buildinpublic
Which is more UX friendly?
Top version - collapsible sidebar menu
Bottom version - dropdown menu
#buildinpublic
Made improvements to the UX selection around agent outputs in a chat interface.
How does this look?
How does this look?
May 6, 2025 at 8:12 AM
Made improvements to the UX selection around agent outputs in a chat interface.
How does this look?
How does this look?
Through talking to users, it became clear that a chat interface doesn’t make site-building intuitive for non-technical folks.
Dev terms aren't part of a layperson's vocab. That’s why we're releasing Specifications Mode, where a PM agent translates your ideas into specs.
Dev terms aren't part of a layperson's vocab. That’s why we're releasing Specifications Mode, where a PM agent translates your ideas into specs.
May 6, 2025 at 3:12 AM
Through talking to users, it became clear that a chat interface doesn’t make site-building intuitive for non-technical folks.
Dev terms aren't part of a layperson's vocab. That’s why we're releasing Specifications Mode, where a PM agent translates your ideas into specs.
Dev terms aren't part of a layperson's vocab. That’s why we're releasing Specifications Mode, where a PM agent translates your ideas into specs.
Just released Secrets Management in Zapi:
- Build integrations safely without exposing sensitive configurations to AI or in the codebase
- Use a separate set of secrets for preview vs live environments
- Build integrations safely without exposing sensitive configurations to AI or in the codebase
- Use a separate set of secrets for preview vs live environments
April 29, 2025 at 7:16 AM
Just released Secrets Management in Zapi:
- Build integrations safely without exposing sensitive configurations to AI or in the codebase
- Use a separate set of secrets for preview vs live environments
- Build integrations safely without exposing sensitive configurations to AI or in the codebase
- Use a separate set of secrets for preview vs live environments
A changelog is a great forcing function to stay accountable.
Each week at Zapi (heyzapi.com), we:
1️⃣ Prioritize changes that move the needle
2️⃣ Build
3️⃣ Publish updates
If we didn't make a good update or missed one - we reflect and do better.
What's your forcing function?
Each week at Zapi (heyzapi.com), we:
1️⃣ Prioritize changes that move the needle
2️⃣ Build
3️⃣ Publish updates
If we didn't make a good update or missed one - we reflect and do better.
What's your forcing function?
April 21, 2025 at 8:30 AM
A changelog is a great forcing function to stay accountable.
Each week at Zapi (heyzapi.com), we:
1️⃣ Prioritize changes that move the needle
2️⃣ Build
3️⃣ Publish updates
If we didn't make a good update or missed one - we reflect and do better.
What's your forcing function?
Each week at Zapi (heyzapi.com), we:
1️⃣ Prioritize changes that move the needle
2️⃣ Build
3️⃣ Publish updates
If we didn't make a good update or missed one - we reflect and do better.
What's your forcing function?
Love @posthog.com as a brand and product. These days, it is so rare to see truth in advertising.
April 9, 2025 at 2:23 AM
Love @posthog.com as a brand and product. These days, it is so rare to see truth in advertising.
Vibe coding sucks for implementing editors.
We used MDXEditor which is built on top of CodeMirror, Lexical and MDAST.
Cursor Agent tried implementing it and went into a death loop trying to fix its own implementation over and over again. Gemini Pro did no better.
We used MDXEditor which is built on top of CodeMirror, Lexical and MDAST.
Cursor Agent tried implementing it and went into a death loop trying to fix its own implementation over and over again. Gemini Pro did no better.
Shipped a markdown editor in Zapi. No more prompting to edit content.
April 8, 2025 at 9:50 AM
Vibe coding sucks for implementing editors.
We used MDXEditor which is built on top of CodeMirror, Lexical and MDAST.
Cursor Agent tried implementing it and went into a death loop trying to fix its own implementation over and over again. Gemini Pro did no better.
We used MDXEditor which is built on top of CodeMirror, Lexical and MDAST.
Cursor Agent tried implementing it and went into a death loop trying to fix its own implementation over and over again. Gemini Pro did no better.
Shipped a markdown editor in Zapi. No more prompting to edit content.
April 8, 2025 at 9:37 AM
Shipped a markdown editor in Zapi. No more prompting to edit content.
Seems like LLMs also suffer from the Justin Case syndrome when left to their own devices. iykyk
April 8, 2025 at 8:08 AM
Seems like LLMs also suffer from the Justin Case syndrome when left to their own devices. iykyk
Wow! Super blown away by animejs docs. Using animation as nav is so intuitive and so good!
April 4, 2025 at 6:02 AM
Wow! Super blown away by animejs docs. Using animation as nav is so intuitive and so good!
Just launched the ability to build backend capabilities in Zapi! For now, it works great with building forms and waitlists. We're going to be tuning this further over the next couple of weeks so that people can build apps.
Give it a try and let me know your feedback.
Give it a try and let me know your feedback.
April 3, 2025 at 2:33 AM
Just launched the ability to build backend capabilities in Zapi! For now, it works great with building forms and waitlists. We're going to be tuning this further over the next couple of weeks so that people can build apps.
Give it a try and let me know your feedback.
Give it a try and let me know your feedback.
6 months ago, 4o was my daily driver. Now I use >5 models for work:
GPT 4o: Image gen
GPT 4.5: Wordsmith posts
O1, Canvas: Write T&Cs
O3-mini: New default
O3-mini (high): Architect/review complex code
Claude 3.5/3.7: Code in Cursor
What’s in your toolkit? What other models should I try?
GPT 4o: Image gen
GPT 4.5: Wordsmith posts
O1, Canvas: Write T&Cs
O3-mini: New default
O3-mini (high): Architect/review complex code
Claude 3.5/3.7: Code in Cursor
What’s in your toolkit? What other models should I try?
April 1, 2025 at 2:09 AM
6 months ago, 4o was my daily driver. Now I use >5 models for work:
GPT 4o: Image gen
GPT 4.5: Wordsmith posts
O1, Canvas: Write T&Cs
O3-mini: New default
O3-mini (high): Architect/review complex code
Claude 3.5/3.7: Code in Cursor
What’s in your toolkit? What other models should I try?
GPT 4o: Image gen
GPT 4.5: Wordsmith posts
O1, Canvas: Write T&Cs
O3-mini: New default
O3-mini (high): Architect/review complex code
Claude 3.5/3.7: Code in Cursor
What’s in your toolkit? What other models should I try?
Gemini Pro 2.5 is here, and Google isn’t messing around—it beats Claude 3.7 in Aider's Polyglot benchmarks! Can't wait to test how well it works with Zapi(heyzapi.com). But first... let me ship multi-agents.
Benchmarks here: aider.chat/docs/leaderb...
Benchmarks here: aider.chat/docs/leaderb...
March 26, 2025 at 12:49 AM
Gemini Pro 2.5 is here, and Google isn’t messing around—it beats Claude 3.7 in Aider's Polyglot benchmarks! Can't wait to test how well it works with Zapi(heyzapi.com). But first... let me ship multi-agents.
Benchmarks here: aider.chat/docs/leaderb...
Benchmarks here: aider.chat/docs/leaderb...
If your “Vibe Coding” session doesn’t include actual coding—you’re probably “Blind Coding.”
March 25, 2025 at 12:41 PM
If your “Vibe Coding” session doesn’t include actual coding—you’re probably “Blind Coding.”
Somewhere in Claude 3.7 system prompt, there must be something that says: “You’re paid per token—earn your keep!”
March 25, 2025 at 6:21 AM
Somewhere in Claude 3.7 system prompt, there must be something that says: “You’re paid per token—earn your keep!”
Single-agent: evals felt optional.
Multi-agents: evals feel absolutely essential. Feels like I'm flying blind without.
Multi-agents: evals feel absolutely essential. Feels like I'm flying blind without.
March 25, 2025 at 6:13 AM
Single-agent: evals felt optional.
Multi-agents: evals feel absolutely essential. Feels like I'm flying blind without.
Multi-agents: evals feel absolutely essential. Feels like I'm flying blind without.
AI keeps surprising me - when it works, it's like magic. I'm upgrading our single agent architecture for code-gen into a multi-agent setup. I'm only halfway but the results are already blowing me away. Here's a site that Zapi built with a design that I could have never ever done myself. Insane...
March 24, 2025 at 6:19 AM
AI keeps surprising me - when it works, it's like magic. I'm upgrading our single agent architecture for code-gen into a multi-agent setup. I'm only halfway but the results are already blowing me away. Here's a site that Zapi built with a design that I could have never ever done myself. Insane...