Shawn Simister
@narphorium.com
Building AI powered tools to augment human creativity and problem solving in San Francisco. Previously @GitHub Copilot, @Google, 🇨🇦
narphorium.com
narphorium.com
Pinned
Shawn Simister
@narphorium.com
· Oct 23
Structured Prompting | Shawn Simister
Organizing Complexity in the Age of Language Models
narphorium.com
Recent developments like subagents and agent planning tools have got me thinking: what are the right abstractions for natural language programming?
I wrote about how we're rediscovering structured programming but this time for prompts instead of code.
I wrote about how we're rediscovering structured programming but this time for prompts instead of code.
New flavor of imposter syndrome: I haven’t even used all my Claude Code quota today
January 15, 2026 at 9:36 AM
New flavor of imposter syndrome: I haven’t even used all my Claude Code quota today
5 years ago I switched to an ultrawide computer monitor and it’s been one of the best decisions I made. So much of my workflow these days starts as a single thread and then expands into 4 or 5 columns of side-by-side documents, AI agents and custom tools…
January 13, 2026 at 1:43 AM
5 years ago I switched to an ultrawide computer monitor and it’s been one of the best decisions I made. So much of my workflow these days starts as a single thread and then expands into 4 or 5 columns of side-by-side documents, AI agents and custom tools…
Reposted by Shawn Simister
prototyping co-drawing with Gemini Flash 3 at Google
in these demos "thinking" is disabled, which makes the model return tokens very quickly (all videos are realtime), and I find these rapid responses pretty good for the use-cases I'm experimenting with, like:
executing simple diagrams ...
in these demos "thinking" is disabled, which makes the model return tokens very quickly (all videos are realtime), and I find these rapid responses pretty good for the use-cases I'm experimenting with, like:
executing simple diagrams ...
January 12, 2026 at 1:04 PM
prototyping co-drawing with Gemini Flash 3 at Google
in these demos "thinking" is disabled, which makes the model return tokens very quickly (all videos are realtime), and I find these rapid responses pretty good for the use-cases I'm experimenting with, like:
executing simple diagrams ...
in these demos "thinking" is disabled, which makes the model return tokens very quickly (all videos are realtime), and I find these rapid responses pretty good for the use-cases I'm experimenting with, like:
executing simple diagrams ...
Reposted by Shawn Simister
Wrote a blog post about the token encoding format in GPT tokenizers.
Have you ever inspected a tokenizer and seen tokens like оже or ĠнÑĥжно? These are tokens in a custom encoding format used for serialization of byte-level tokenizers.
This post shows how to recover the original token.
Have you ever inspected a tokenizer and seen tokens like оже or ĠнÑĥжно? These are tokens in a custom encoding format used for serialization of byte-level tokenizers.
This post shows how to recover the original token.
Decoding GPT-Style Tokens
Decoding the GPT-style token representation.
mcognetta.github.io
January 5, 2026 at 4:09 PM
Wrote a blog post about the token encoding format in GPT tokenizers.
Have you ever inspected a tokenizer and seen tokens like оже or ĠнÑĥжно? These are tokens in a custom encoding format used for serialization of byte-level tokenizers.
This post shows how to recover the original token.
Have you ever inspected a tokenizer and seen tokens like оже or ĠнÑĥжно? These are tokens in a custom encoding format used for serialization of byte-level tokenizers.
This post shows how to recover the original token.
Reposted by Shawn Simister
Introducing Splat 🟨📌: Affinity diagramming in a single HTML file. Drag notes and organize ideas on a visual board. With semantic search and an agentic AI assistant. Open source and fully local. Check it out here: github.com/ianarawjo/sp...
December 30, 2025 at 5:49 PM
Introducing Splat 🟨📌: Affinity diagramming in a single HTML file. Drag notes and organize ideas on a visual board. With semantic search and an agentic AI assistant. Open source and fully local. Check it out here: github.com/ianarawjo/sp...
“If understanding lives in prompts and specifications rather than implementations, what happens to the craft of programming? … The definition of quality may be moving from "durability" to "regenerability"—from code that lasts to code that can be reliably reproduced.”
December 27, 2025 at 7:42 PM
“If understanding lives in prompts and specifications rather than implementations, what happens to the craft of programming? … The definition of quality may be moving from "durability" to "regenerability"—from code that lasts to code that can be reliably reproduced.”
Love this vision for sensemaking computers
Telepath @telepathinc.bsky.social (Stephen Hood, Josh Whiting and @ruperts.world , telepath.computer) is the first sensemaking computer: a revolutionary new personal computing experience, built from the ground up for the AI era.
www.youtube.com/watch?v=jEm5...
www.youtube.com/watch?v=jEm5...
Telepath's Sensemaking Computer Upends Personal Computing
YouTube video by Betaworks
www.youtube.com
December 9, 2025 at 12:56 AM
Love this vision for sensemaking computers
Claude Code is starting to write out functional requirements as part of the planning process
www.anthropic.com/engineering/...
www.anthropic.com/engineering/...
November 27, 2025 at 8:53 PM
Claude Code is starting to write out functional requirements as part of the planning process
www.anthropic.com/engineering/...
www.anthropic.com/engineering/...
Reposted by Shawn Simister
FIles are the new database -- most codebases are already well structured, why not aid the coding agent with a file based index at the root of each folder, generated in seconds using TreeSitter? Runs as a skill, invoked on demand. Supports JS, TS, Ruby, Go, Rust, Java, and of course Python.
Code Mapping for AI-Assisted Development | Oskar | WhiteWind blog
Code Mapping for AI-Assisted Development
The Problem
When working with unfamiliar codebases, Claude needs context about code structure before making changes. Reading every file wastes tokens and time. Asking Claude to "explore the codebase" produces inconsistent results and burns through context w...
whtwnd.com
November 27, 2025 at 8:43 PM
FIles are the new database -- most codebases are already well structured, why not aid the coding agent with a file based index at the root of each folder, generated in seconds using TreeSitter? Runs as a skill, invoked on demand. Supports JS, TS, Ruby, Go, Rust, Java, and of course Python.
Putting on my noise cancelling headphones so I can listen to some high quality noise
November 12, 2025 at 1:52 AM
Putting on my noise cancelling headphones so I can listen to some high quality noise
I’m seeing confusion about Claude Skills and why you need them. It sort of feels like Anthropic is building tools that fix problems many people won’t know that they have until next year
www.anthropic.com/engineering/...
www.anthropic.com/engineering/...
November 10, 2025 at 1:26 AM
I’m seeing confusion about Claude Skills and why you need them. It sort of feels like Anthropic is building tools that fix problems many people won’t know that they have until next year
www.anthropic.com/engineering/...
www.anthropic.com/engineering/...
I’m seeing a lot of AI coding tools trying to figure out multi-agent UX. To me, delegating feels like the least interesting part of the problem…
October 30, 2025 at 2:28 AM
I’m seeing a lot of AI coding tools trying to figure out multi-agent UX. To me, delegating feels like the least interesting part of the problem…
Reposted by Shawn Simister
Do AI agents ask good questions? We built “Collaborative Battleship” to find out—and discovered that weaker LMs + Bayesian inference can beat GPT-5 at 1% of the cost.
Paper, code & demos: gabegrand.github.io/battleship
Here's what we learned about building rational information-seeking agents... 🧵🔽
Paper, code & demos: gabegrand.github.io/battleship
Here's what we learned about building rational information-seeking agents... 🧵🔽
October 27, 2025 at 7:17 PM
Do AI agents ask good questions? We built “Collaborative Battleship” to find out—and discovered that weaker LMs + Bayesian inference can beat GPT-5 at 1% of the cost.
Paper, code & demos: gabegrand.github.io/battleship
Here's what we learned about building rational information-seeking agents... 🧵🔽
Paper, code & demos: gabegrand.github.io/battleship
Here's what we learned about building rational information-seeking agents... 🧵🔽
Reposted by Shawn Simister
More detail including a code sample for trivially implementing linking in an agent on my blog: mbleigh.dev/posts/co...
Links were powerful enough to build the entire web...I think they'll be critical for building the context for agents as well.
Links were powerful enough to build the entire web...I think they'll be critical for building the context for agents as well.
Context engineering is sleeping on the humble hyperlink
The dream of hypermedia can come to fruition in a world where clients can not only parse, but navigate with intent.
mbleigh.dev
October 23, 2025 at 1:39 PM
More detail including a code sample for trivially implementing linking in an agent on my blog: mbleigh.dev/posts/co...
Links were powerful enough to build the entire web...I think they'll be critical for building the context for agents as well.
Links were powerful enough to build the entire web...I think they'll be critical for building the context for agents as well.
Reposted by Shawn Simister
Discerning what is at the core, and how to communicate it with others, has taken years. This might be the closest I've come so far.
Part 2 of The Interfaces With Which We Think is out now...
Start at the intro: alexanderobenauer.com/think/
Start at Part 2: alexanderobenauer.com/think/2/
Part 2 of The Interfaces With Which We Think is out now...
Start at the intro: alexanderobenauer.com/think/
Start at Part 2: alexanderobenauer.com/think/2/
The Interfaces With Which We Think
The concepts in modern operating systems — apps, windows, desktops, notifications, and so on — have so permeated our understanding of personal computing, it’s hard to imagine anything else, let alone ...
alexanderobenauer.com
October 23, 2025 at 2:26 PM
Discerning what is at the core, and how to communicate it with others, has taken years. This might be the closest I've come so far.
Part 2 of The Interfaces With Which We Think is out now...
Start at the intro: alexanderobenauer.com/think/
Start at Part 2: alexanderobenauer.com/think/2/
Part 2 of The Interfaces With Which We Think is out now...
Start at the intro: alexanderobenauer.com/think/
Start at Part 2: alexanderobenauer.com/think/2/
Recent developments like subagents and agent planning tools have got me thinking: what are the right abstractions for natural language programming?
I wrote about how we're rediscovering structured programming but this time for prompts instead of code.
I wrote about how we're rediscovering structured programming but this time for prompts instead of code.
Structured Prompting | Shawn Simister
Organizing Complexity in the Age of Language Models
narphorium.com
October 23, 2025 at 2:38 AM
Recent developments like subagents and agent planning tools have got me thinking: what are the right abstractions for natural language programming?
I wrote about how we're rediscovering structured programming but this time for prompts instead of code.
I wrote about how we're rediscovering structured programming but this time for prompts instead of code.
Reposted by Shawn Simister
New post 🎉
Going back to my roots on writing about the inner workings of things, a breakdown of key-value databases and how you might make one from scratch:
nan.fyi/database
Going back to my roots on writing about the inner workings of things, a breakdown of key-value databases and how you might make one from scratch:
nan.fyi/database
October 21, 2025 at 3:00 PM
New post 🎉
Going back to my roots on writing about the inner workings of things, a breakdown of key-value databases and how you might make one from scratch:
nan.fyi/database
Going back to my roots on writing about the inner workings of things, a breakdown of key-value databases and how you might make one from scratch:
nan.fyi/database
Reposted by Shawn Simister
An AI coding approach I've liked using, which I first used on this project:
Vibe code the first version. Investigate the code for its shape, complexities, needs, failure points.
Then scrap all that and engineer a new version.
Vibe code the first version. Investigate the code for its shape, complexities, needs, failure points.
Then scrap all that and engineer a new version.
This article is now public! Link in reply.
I recreated Jef Raskin's 1987 'Canon Cat' user environment and observed myself as I lived in it for a week.
Implementation details, surprises, and more in the essay.
(Why touch grass when you can read 3.2k words on user environments?)
I recreated Jef Raskin's 1987 'Canon Cat' user environment and observed myself as I lived in it for a week.
Implementation details, surprises, and more in the essay.
(Why touch grass when you can read 3.2k words on user environments?)
October 20, 2025 at 3:01 PM
An AI coding approach I've liked using, which I first used on this project:
Vibe code the first version. Investigate the code for its shape, complexities, needs, failure points.
Then scrap all that and engineer a new version.
Vibe code the first version. Investigate the code for its shape, complexities, needs, failure points.
Then scrap all that and engineer a new version.
Great analysis of spec-driven development tools and recommendations for how to make them more effective: more variety of workflows, smaller iteration cycles, and better spec review tools
I tried to make sense of "spec-driven development" by looking at 3 tools: Amazon's Kiro, GitHub's spec-kit, and the Tessl Framework
martinfowler.com/articles/exp...
martinfowler.com/articles/exp...
Understanding Spec-Driven-Development: Kiro, spec-kit, and Tessl
Notes from my Thoughtworks colleagues on AI-assisted software delivery
martinfowler.com
October 15, 2025 at 5:41 PM
Great analysis of spec-driven development tools and recommendations for how to make them more effective: more variety of workflows, smaller iteration cycles, and better spec review tools
Reposted by Shawn Simister
Some thoughts on LLMs, Autopilot, and VibeCoding rahulpandita.me/blog/2025-10...
October 13, 2025 at 2:30 AM
Some thoughts on LLMs, Autopilot, and VibeCoding rahulpandita.me/blog/2025-10...
These new Windsurf codemaps look a lot like Project Copernicus from @kcieslak.io
githubnext.com/projects/cop...
githubnext.com/projects/cop...
October 13, 2025 at 1:45 AM
These new Windsurf codemaps look a lot like Project Copernicus from @kcieslak.io
githubnext.com/projects/cop...
githubnext.com/projects/cop...
Reposted by Shawn Simister
6/
That said, HRM and TRM are fascinating proof‑of‑concepts that show what’s possible with relatively small and efficient architectures. I'm still curious what the real‑world use case will look like. Maybe they could serve as reasoning or planning modules within a larger tool‑calling system.
That said, HRM and TRM are fascinating proof‑of‑concepts that show what’s possible with relatively small and efficient architectures. I'm still curious what the real‑world use case will look like. Maybe they could serve as reasoning or planning modules within a larger tool‑calling system.
October 9, 2025 at 4:23 PM
6/
That said, HRM and TRM are fascinating proof‑of‑concepts that show what’s possible with relatively small and efficient architectures. I'm still curious what the real‑world use case will look like. Maybe they could serve as reasoning or planning modules within a larger tool‑calling system.
That said, HRM and TRM are fascinating proof‑of‑concepts that show what’s possible with relatively small and efficient architectures. I'm still curious what the real‑world use case will look like. Maybe they could serve as reasoning or planning modules within a larger tool‑calling system.
Reposted by Shawn Simister
We have 🤖 AI notetakers in meetings but continue to silo know-how every time we close terminals. Not just the how but also the why and what.
Sign up for beta access to visr.sh: it's like as if Granola AI and #tmux had a kid.
Sign up for beta access to visr.sh: it's like as if Granola AI and #tmux had a kid.
Visr: The Agentic Terminal Notepad Integral to Your Docs
The Agentic Terminal Notepad Integral to Your Docs
visr.sh
October 7, 2025 at 5:05 PM
Fun panel discussion on spec-based development with @kiro.dev team.
Especially interested in upcoming features to ground the specs to verify that the work is done correctly
Especially interested in upcoming features to ground the specs to verify that the work is done correctly
October 7, 2025 at 3:45 AM
Fun panel discussion on spec-based development with @kiro.dev team.
Especially interested in upcoming features to ground the specs to verify that the work is done correctly
Especially interested in upcoming features to ground the specs to verify that the work is done correctly
Reposted by Shawn Simister
What, if anything, do LLMs change about interface design?
Interfaces are languages
Is there a universal translation?
shimmeringvoid.leaflet.pub
October 4, 2025 at 2:56 AM
What, if anything, do LLMs change about interface design?