iiinova — where your digital self keeps evolving

Reposted by Jon Walls

Tobias Wilson-Bates @phdhurtbrain.bsky.social · 21/08/2026

James Bond: you think I haven’t faced death before? Supervillain: oh, I don’t expect you to die, Mr. Bond. I expect you to suffer. *opens travel funding reimbursement software I think you will find your motorboat rental and casino reservations especially complex James Bond: no. no. pls kill me

382280469

Jon Walls @jonwalls.dev · 14/06/2026

I have learnt that when my friendly little coding agent says this, it can often translate to "that's a damned stupid idea but I'll do it anyway."

Message from Claude Code: "Your reframing is exactly the cleanest architecture — and it works."

010

Reposted by Jon Walls

warrior cop @wyattprivilege.bsky.social · 19/05/2026

I GET LOGGED OUT. BUT I LOG IN AGAIN

149155833098

Jon Walls @jonwalls.dev · 18/05/2026

Not gonna lie, I find this difficult to read... (strange rendering issues for Claude Code inside VS Code)

010

Jon Walls @jonwalls.dev · 04/05/2026

I like the subtlety of leaving the “Why?” section blank.

100

Jon Walls @jonwalls.dev · 01/04/2026

So is Claude's new buddy thing a random pet or a personality profile of the user? "No snark but also no patience" is probably a fair assessment of my interactions with Claude! I'm guessing released since letting their source code into the wild must have accelerated a few decisions!

000

Jon Walls @jonwalls.dev · 19/03/2026

This is such a beautiful little idea. Wonderful!

010

Jon Walls @jonwalls.dev · 17/03/2026

It's one of the new phrases of the age, isn't it? (working with coding agents)

Message from a coding agent: "You're right - I went overboard"

001

Jon Walls @jonwalls.dev · 12/03/2026

Isn’t that also a profound misunderstanding of what LLMs do? Especially when “confidently wrong” and “sycophancy” have been widely recognised challenges since they first came to broad attention?

161

Jon Walls @jonwalls.dev · 07/03/2026

There's some great terminology in deep learning. I will never again make a wild-ass guess. From now on, I only make high-variance Monte Carlo estimates.

011

Jon Walls @jonwalls.dev · 07/03/2026

Pity the poor bastard whose dinghy is at those coordinates

000

Jon Walls @jonwalls.dev · 05/03/2026

Hey @anthropic.com please don't do a Microsoft and make it impossible to figure out how to just login. I just want to try out your voice mode in Claude Code. I've tried all the Anthropic options via /login. Do you have a third hidden option? Is claude.ai considered a third party service?

Screencap showing login options for Claude Code:

Claude Code can be used with your Claude subscription or billed based on API usage through your Console account.

Select login method:

❯ 1. Claude account with subscription · Pro, Max, Team, or Enterprise
2. Anthropic Console account · API usage billing
3. 3rd-party platform · Amazon Bedrock, Microsoft Foundry, or Vertex AI

000

Jon Walls @jonwalls.dev · 03/03/2026

This is something I feel deserves more mindshare - AI is good with code, but that can mean using just a language and its standard library, adding a library, using an SDK, a DSL, an API call, etc.

000

Jon Walls @jonwalls.dev · 11/02/2026

My attempts ("Omcrn" etc) always have it going through at least a short loop as soon as it generates "Omicron" as the target answer. I really need to get back to work :D but will be keeping an eye out for explanations to this!

100

Jon Walls @jonwalls.dev · 11/02/2026

altho I see counting in your original as well, mine is similar but slight different wording. Very strange. For added weirdness, I tried adding a typo (was curious if maybe tokenization related with omicron as a "special character")

100

Jon Walls @jonwalls.dev · 11/02/2026

So it is! Slightly different tho, guess even non-extended has a layer over the model's thinking? Something awry with the "counting the r's in strawberry" fix? The trace is "recounting": "The user wrote "Omicron" which is missing the 'i' — no wait, let me recount: O-m-i-c-r-o-n vs O-m-i-c-r-o-n."

110

Jon Walls @jonwalls.dev · 11/02/2026

Interesting! This happens for me in Opus 4.6, but not with Extended Thinking turned on.

120

Reposted by Jon Walls

shri @shrikhalpada.dev · 06/02/2026

maybe my favorite all time correlation

a chart showing a search trend spike every year for "roman numerals" around the super bowl

36719

Jon Walls @jonwalls.dev · 07/02/2026

We’re not remotely close to the point where AI is conscious. The difference matters. Imagine a provocateur sending out bots specifically to provoke people into using slurs, against these machines built only to harm them. Many would take your piece as an argument for going after those people, hard.

000

Jon Walls @jonwalls.dev · 07/02/2026

Very interesting. The core point I like is that how people behave in public, how we respond, is important and impactful. It matters. Where I disagree is the idea sentience, experiencing pain or cruelty, doesn’t matter. We mustn’t invent yet more reasons to judge people for their private behaviour.

130

Reposted by Jon Walls

Rill Data @rilldata.com · 04/02/2026

.@medriscoll.com was on The Joe Reis podcast to discuss the evolving landscape of business intelligence and data engineering. They explore why the industry keeps "rediscovering" old concepts like the semantic layer and how the rise of AI agents is forcing us to rethink how we structure data.

102

Jon Walls @jonwalls.dev · 04/02/2026

This is super interesting. Go is a lovely language to work with, and compiled binaries offer a very nice experience to what we so often experience in Python land!

000

Reposted by Jon Walls

Uncle Duke @uncleduke1969.bsky.social · 31/01/2026

the theater is screening DOS Boot

Electronic billboard contains a fragment of an old Intel boot screen.

361777363

Jon Walls @jonwalls.dev · 14/01/2026

The big one: does the company realise AI can’t think for you?

000

Jon Walls @jonwalls.dev · 14/01/2026

Like most mandates. Bad managers can turn good ideas into a bad ones, but you can’t judge on that basis. Does it mean the devs get their AI paid for? Mandates can stop bad managers blocking their staff. Does it allow for exceptions? Does it differentiate ideation, prototyping, coding, reviews?

110

Jon Walls @jonwalls.dev · 07/01/2026

Great article. What LLMs mean for education is a big question. My personal analogy is that Christopher Nolan studied English: learning how to think about a script is as important as the skills a director needs to make a movie (if even someone/something else does the work). Same for CS vs coding.

010

Jon Walls @jonwalls.dev · 06/01/2026

So true. Thinking first in terms of the tasks you want performed, before thinking about the agents you want to build, is a good order to go in.

010

Jon Walls @jonwalls.dev · 05/01/2026

I love the smell of burning tokens in the morning. (This is the new motto of everyone leaning into agentic coding)

000

Jon Walls @jonwalls.dev · 03/01/2026

Nice first test run of Beads (github.com/steveyegge/b...) from @steve-yegge.bsky.social. A new tool slotting smoothly into an existing project is nice. It really jibes with my way of thinking to have the coding agent use a real-but-lightweight issue tracker. Good summary from Claude on completion.

010

Jon Walls @jonwalls.dev · 02/01/2026

I do love a good data vis labour of love

100

Jon Walls @jonwalls.dev · 01/01/2026

Designing Data-Intensive Applications was an instant classic - nice to know 2nd ed. is coming!

010

Reposted by Jon Walls

Rill Data @rilldata.com · 19/12/2025

“Sometimes the best reward isn't money. It's watching the giants scramble to copy your stupid SQL feature." - Hannes Mühleisen, DuckDB In the updated episode, we discussed: → Why they worked in secret for a year → Why 99% of analytics workloads don't need distributed systems datatalks.rilldata.com

011

Jon Walls @jonwalls.dev · 03/11/2025

Liked this post by Ahmad El-Alfy (via HNews thread) on URLs as state containers. Often neglected in app design but compelling to end users. Not so easy to get right since not all state should be in URL, and weirdness can happen with permissions. Always worth it though. alfy.blog/2025/10/31/y...

alfy.blog

Your URL Is Your State

A deep dive into how thoughtful URL design can enhance usability, shareability, and performance. Learn what state belongs in URLs, common pitfalls to avoid, and practical patterns for modern web apps.

010

Reposted by Jon Walls

Elle @elleisanisland.bsky.social · 12/09/2025

All I saw was cheesecake boat and ocean for at least 5 minutes

Photo of a missing slice of cheesecake that looks like a cheesecake hull floating in a deep blue ocean. The "ocean" is actually the blueberry topping on the large cheesecake that we don't see completely.

419244895353

Reposted by Jon Walls

Rill Data @rilldata.com · 17/07/2025

Welcome to our 💡Tip Thursday💡 series where @jonwalls.dev highlights ways to slice & dice your data. The goal of each Rill dashboard is to provide users with metrics & dimensions for various use cases and to create an interactive experience to cut data in any form. www.youtube.com/playlist?lis...

youtube.com

Exploring Data with Rill - YouTube

Series of short tutorials showing you how to slice'n'dice your data in Rill

101

Jon Walls @jonwalls.dev · 08/06/2025

Enjoying messing around with Claude Code! It does say "you're absolutely right" a lot, though. Pretty sure I'd have been done with my little weekend project by now if I was "absolutely right" with every idea and prompt...

000

Jon Walls @jonwalls.dev · 02/06/2025

My (limited!) experience is - Forgive yourself your first recordings. You’ll get better, but first you must ship. - You do get used to yourself. Enough so you’re focused on what you’re trying to achieve. - The only criticism that really hurts is of your ideas. Same risk as writing them down!

020

Jon Walls @jonwalls.dev · 02/06/2025

The performance and editing are skills for sure. “Artifice” implies dishonesty but it’s like learning an instrument: authentic performance takes lots of practice. Without practice, the audience pays more attention to the quality of the performance than anything you actually say.

110

Jon Walls @jonwalls.dev · 31/05/2025

"You're absolutely right on all counts." Oh LLM sycophants, I may talk like it (it's a professional hazard in tech), but it really isn't true.

000

Jon Walls @jonwalls.dev · 29/05/2025

A bit inspired by @simonwillison.net 's stream of vibe-coded "low stakes" tools, was pleased with Claude's take on explaining composition effects. The ratio of a subset of unique counts to the total unique count falls the longer the time window. Which still boggles my mind. How can 7 x 29 = 6?!

Data tool demonstrating how calculating ratios of unique counts can counter-intuitively result in much lower ratios over 7 days versus 1 day.

000

Jon Walls @jonwalls.dev · 30/04/2025

Good policy. LLMs are useful, but have also provided plenty of examples of how knowing when to say “no” is important!

020

Reposted by Jon Walls

John Sabath @johnsabath.com · 07/04/2025

@dazzle.fm's internal topic heatmap for Bluesky is now available to the public dazzle.fm/stats Seeing surges of discussion around a topic and working backwards to figure out what caused the spike is a fun past time The 1am science spike is still a mystery though

1368

Jon Walls @jonwalls.dev · 03/04/2025

That's one helluva landing page!

000

Jon Walls @jonwalls.dev · 09/03/2025

Sixteenth century version of having too many tabs open

000

Reposted by Jon Walls

rmoff 🏃‍♂️🫖🥓 @rmoff.net · 28/02/2025

Blogged: Exploring UK Environment Agency data with @duckdb.org and @rilldata.com rmoff.net/2025/02/28/e... #dataBS

0207

Jon Walls @jonwalls.dev · 16/02/2025

Work has had me in JS world for years and recently Go. I’ll still fire up the occasional Python notebook to try out a technique or perform a quick task. Tried uv recently when I got a new machine, and was gearing up to review my Python setup. It’s sooooo much better. Good infra. Recommended.

000

Jon Walls @jonwalls.dev · 14/02/2025

Scenarios where including a viz guru is better/faster: cheaper: - data needs wrangling - viz work is 10x faster with data tools - most data does not “speak” w/o normalisation, or variance, or cohorting, or … - viz gurus make *interactive* tools - can export viz to vector formats for design collab.

120

Jon Walls @jonwalls.dev · 14/02/2025

I’m in the camp that data vis is a great skill, have spent lots of time with d3. But can’t think of any full time, long term, inhouse data viz specialists. Except maybe outsourced labour churning out dashboards. Isn’t good viz work better paid as design, analysis, engineering, strategy, etc?

010