Sign in

iiinova is part of the Atmosphere — an open, public network where you own your identity.

AJ Fisher

@ajfisher.social
115 followers 206 following 361 posts

Creator of experiences combining web, AI/ML, hardware & data - sometimes they work. Dev, interaction researcher, tinkerer. CTO@LoyPal & Rocket Melb / 🌐 ajfisher.me / was @ajfisher on 🐦

PostsRepliesMedia
AJ Fisher @ajfisher.social · 24/09/2026
💯 I use this on my personal site deliver markdown when it's an obvious agent request and it's dead simple for simple blog content. Complex, referential data that's auditable steps the complexity up a couple of notches. There is an MD version of the report but references are less rich.
000 0
AJ Fisher @ajfisher.social · 24/09/2026
Responsive design is taking a new turn when you consider agentic UX. I've pulled together some thoughts based on a recent project around information design, managing provenance, managing trust and building content verification systems. #ai #ux #agents ajfisher.me/2026/09/24/i...
ajfisher.me
Information design for the agentic age
Most information on the web is published primarily for humans, even though increasingly the thing consuming it is likely an agent. We built a version of our annual report explicitly for agents which f...
120 2
AJ Fisher @ajfisher.social · 28/07/2026
For those that don't fancy a full read - here's a little infographic summary that you might want to peruse instead to whet your appetite. #aiFuture #ai
An infographic comparing two potential stabe AI futures - a controlled equilibrium or an adaptive equilibrium.

Both have different features on how to get there but run risks to how durable they are.
000 0
AJ Fisher @ajfisher.social · 28/07/2026
📖 AI has only two stable futures: a controlled equilibrium built on coordinated restraint, or an adaptive equilibrium built after capability diffuses. Nuclear deterrence is the wrong analogy. Montreal may be closer, but geopolitics makes control fragile. ajfisher.me/2026/07/28/t... #AIGovernance
ajfisher.me
AI has only two stable futures
Two futures become apparent once frontier AI becomes powerful enough to reshape society. One depends on containing its diffusion. The other depends on adapting to it.
100 1
AJ Fisher @ajfisher.social · 14/07/2026
If AI keeps advancing but most of us only get a flatter, more constrained version, are today’s models already good enough for much of commercial work? If so, advantage comes from the grindy work of closing the deployment gap, not waiting for GPT-6. ajfisher.me/2026/07/14/a... #AI #EnterpriseAI
ajfisher.me
What happens if this is as good as it gets for AI?
For many routine commercial tasks, model capability may no longer be the constraint. The goal is to close the deployment gap and turn existing AI into realised economic value.
000 0
AJ Fisher @ajfisher.social · 24/03/2026
Agents building agent swarms that are using sub agents to build software... abstraction is getting bananas.... #aiswe #agents #coding
020 2
AJ Fisher @ajfisher.social · 23/03/2026
I don't believe so, or at least not yet.
000 0
AJ Fisher @ajfisher.social · 23/03/2026
I didn't test it extensively though because it didn't run fast enough (was running locally on my MBP) to be really viable. Ie something that takes a minute to get through 1 cycle wasn't awesome so I just sort of abandoned it. But I reckon Qw3.5 or thereabouts might be close to the minimum...
100 1
AJ Fisher @ajfisher.social · 23/03/2026
100% agree. There's a minimal level of competance that is needed for a harness to actually be effective. I did a test a while back when Qwen 3.5 released and even a quantised model (27B) "sort of worked" - I'd suspect this would be close to the minimal bar.
100 1
AJ Fisher @ajfisher.social · 23/03/2026
I will give this a go and report back. TBH at this point at lot of this is hypothetical as I tested the broad principles but need to develop the loop further. Having exit criteria on stuck is a good idea though - will design a test for this and see what the right scaffolding looks like.
100 1
AJ Fisher @ajfisher.social · 14/03/2026
All the powers of these AI tools - yet none of them can find me the exact replacement bolt I need for a chair. This meat sack still has some use after all... #ai #humanSkills
000 0
AJ Fisher @ajfisher.social · 27/02/2026
After doing some napkin math(s) I noticed that a marginally competent but high-velocity model could out perform a slower, high-performance SOTA model via sheer iteration speed. This is exactly the scenario that underpins most agentic coding loops so could have some large impacts.
100 1
AJ Fisher @ajfisher.social · 27/02/2026
This week a new language model launched: Mercury 2 by inception labs. It's really, really fast & shows the potential that diffusion language models could have. I knocked together a piece last night, that works through some of the implications. ajfisher.me/2026/02/26/m... #llm #diffusion #agents
ajfisher.me
Mercury 2 won’t outthink frontier models but diffusion might out-iterate them
The release of Mercury 2 suggests diffusion models may change agent design - not by increasing intelligence, but by making iteration dramatically faster. As iteration gets cheaper, verification become...
100 1
AJ Fisher @ajfisher.social · 06/02/2026
Australian political parties are not meeting the needs of the moment and it's time for a reset. This widening rift could actually be a gift for the next generation of political leaders and candidates.
010 1
AJ Fisher @ajfisher.social · 06/02/2026
Ultimately, this could create a genuine centre right party (a stack of teal independents does not a party make) that provides a credible opposition (or govt) and extricates then from the mess that is caused by the LNP / Nat fringe dwellers trying to out- Hanson PH.
100 1
AJ Fisher @ajfisher.social · 06/02/2026
With preferential voting, preferences will flow between the two parties anyway due to their natural conservative base so the votes will still aggregate that way first. The same way many people vote for the Greens then they slide towards Labor during elimination.
100 1
AJ Fisher @ajfisher.social · 06/02/2026
Fielding competitive candidates would also force the Nats to shape up, get eaten by PHON or become a small, largely irrelevant minor party. Maybe they end up like the Greens, representing a narrow slice of the electorate from a specific perspective.
100 1
AJ Fisher @ajfisher.social · 06/02/2026
Duts sold it, but outside the Nats and QLD LNP, no one was really committed to the nuclear plan. It was a joke they were forced to sell and the electorate could see the dysfunction. A cohesive, centre right policy agenda could arrest the rot at the heart of the coalition.
100 1
AJ Fisher @ajfisher.social · 06/02/2026
The Libs could credibly (outside of QLD) field candidates in most electorates in the country. Unshackled from the tail wagging Nats and community annoyance, they would naturally move towards the centre because that would give them the best chance to be elected on *a single policy platform*
100 1
AJ Fisher @ajfisher.social · 06/02/2026
The coalition agreement means libs and Nats don't run two candidates in most seats. This is really only to be efficient with their campaign spend. But they have plenty of money. Fundraising isn't really the issue if you look at the hundreds of millions of dollars spent at the last election.
100 1
AJ Fisher @ajfisher.social · 06/02/2026
The Coalition isn't winning the next federal election anyway (barring some form of Labor implosion). The Libs and Nats have increasingly different positions on too many issues. Split, go through the cycle and re-evaluate after. It might be for the better.... 🧵 #auspol
100 1
AJ Fisher @ajfisher.social · 03/02/2026
You also don't need to give it full access to your entire digital world to have an impact. Asnwering emails is a terrible thing for an orchestration agent to do (and Gemini / Copilot can do this already). Give it some meatier tasks instead.
000 0
AJ Fisher @ajfisher.social · 03/02/2026
Like most of the dev community the last couple of weeks, I have been playing with clawdbot / moltbot / openclaw. You don't need crazy hardware to run it. A simple Raspberry Pi is plenty enough to be useful. I did a write up of my setup here: ajfisher.me/2026/02/03/o... #clawdbot #openclaw #ai
ajfisher.me
No Mac Mini, no worries: running OpenClaw on a Raspberry Pi.
OpenClaw gives us a taste of a personal AI assistant that can actually do things, but you don't need to buy an expensive Mac Mini or give full access to your digital life for it to be useful. This gui...
100 1
Reposted by AJ Fisher
Tynan is afk til 10/19 ⁂ @tynanpurdy.com · 20/01/2026
I have no more patience for platforms. I'm done. #offprint tynanistyping.offprint.app/a/3mcsvj…
tynanistyping.offprint.app
Tynan Is Typing | Tynan Purdy | Offprint
Take a peek inside my brain. Internet thoughts. Design takes. Pontificating on life.
512328 123
AJ Fisher @ajfisher.social · 20/01/2026
Also if you're curious, yep I did throw a coding agent at this and results were mixed-good. Net beneficial once I had the groundwork / basic plumbing in place. Because I created a new site in my repo alignside my old one, there was a little bit of confusion from time to time.
000 0
AJ Fisher @ajfisher.social · 20/01/2026
If you're curious - I switched to Astro (astro.build) which is a delight to use, incredibly fast to build (thanks Vite), super lightweight and allows you to just work with basic HTML building blocks + a bit of logic to hook them together into templates. Highly recommended.
astro.build
Astro
Astro builds fast content sites, powerful web applications, dynamic server APIs, and everything in-between.
100 1
AJ Fisher @ajfisher.social · 20/01/2026
After the disaster that has been Gatsby after the Netlify acquisition, have finally managed to move ajfisher.me to be more simple, ditch React, and be JS free. So I'm assuming this makes my rants / musings 100% certified organic or something now?? #website #blog #development
ajfisher.me
Home | ajfisher
Explore insights on technology, business and user experience at the intersection of AI, web, media and digital innovation.
110 1
AJ Fisher @ajfisher.social · 27/12/2025
We'll know who has won and lost this game by 3pm today. #AUSvENG #ashes
000 0
AJ Fisher @ajfisher.social · 26/12/2025
No. Was a bit too late on the tickets. But rewatched at home
000 0
AJ Fisher @ajfisher.social · 26/12/2025
Can't believe how many wickets today. Incredible bowling pitch
100 1
AJ Fisher @ajfisher.social · 26/12/2025
Scott Boland, OAM.... #AUSvENG #ASHES
110 1
AJ Fisher @ajfisher.social · 22/12/2025
Some other teaser predictions: - HR will be sitting in the hotseat during board meetings - Disposable Software arrives that business users will love and IT will hate. - The China manufacturing shock continues but this time with autonomous robots. #hr #ai #work #strategy #business #china #robotics
011 1
AJ Fisher @ajfisher.social · 22/12/2025
As we come into 2026, my hot take is that Replit, Cursor etc get acquired, crushed or sold for parts. Bundling will make a comeback as MS just make "Agentic Edit" part of a standard M365 or Github license. This and more ajfisher.me/2025/12/22/a... #ai #predictions #blog #business
ajfisher.me
AI, incentives and power in 2026
The AI hype is over, and the industrial reality has begun. Here are some predictions for 2026, from the rise of disposable software to potential clashes over worker rights.
101 1
AJ Fisher @ajfisher.social · 20/12/2025
Tell us what you really think of the batting Ricky. #ashes #AUSvENG
000 0
AJ Fisher @ajfisher.social · 20/12/2025
December 20, 2025. Third Test, Adelaide, Day 4 - England 2nd Innings. Mark the day as the final death of the Bazball experiment and England beginning the climb back to test cricket relevance. #ashes #AUSvENG
010 1
AJ Fisher @ajfisher.social · 19/12/2025
Animated ☝️👅👍 all day long. What a future we're barreling towards...
010 1
AJ Fisher @ajfisher.social · 19/12/2025
Now they have digitally mapped signage on the pitch instead of painted sponsorship I'm just waiting for a broadcaster to open up real time bidding for advertising - AdWords for Pitches
110 1
AJ Fisher @ajfisher.social · 19/12/2025
My British grandad brought me one back from the UK when I was seven (can't recall brand now) but had to soak it in oil and everything before using it. Legit cried when I snapped it when I was 10. I did get a nice shiny new Slazenger V something after that though
020 2
AJ Fisher @ajfisher.social · 19/12/2025
Last night I was at the BBL and the number of sponsored "crowd engagement hooks" was out of control. At least it was for benign stuff like broadband rather than sports betting (unlike *coughs* AFL/ NRL)
110 1
AJ Fisher @ajfisher.social · 19/12/2025
Not a super sophisticated web app by any stretch but the fact I did this while sitting at the cricket while on my phone with some occasional steering brings the fun back to making these sorts of things and is something that has sat in my "sort of middle play around with list" for ages.
000 0
AJ Fisher @ajfisher.social · 19/12/2025
Gemini canvas / artifacts / snippets getting better by the week at the moment. Silly little web app sequencer thing. Have a play, make some tunes. gemini.google.com/share/c9c5f9... #ai #genAI #gemini
gemini.google.com
‎Gemini - direct access to Google AI
Created with Gemini
100 1
AJ Fisher @ajfisher.social · 08/12/2025
Codex-coded and human-steered, it was produced on a Saturday afternoon whilst having to do some other tasks that needed much higher attention. You can checkout the code in github and give it a whirl for yourself. github.com/ajfisher/llm... #vibecoded #claude #codex
github.com
GitHub - ajfisher/llm-advisors: Council of advisors using existing tooling
Council of advisors using existing tooling. Contribute to ajfisher/llm-advisors development by creating an account on GitHub.
020 2
AJ Fisher @ajfisher.social · 08/12/2025
Inspired by Andrej Karpathy's LLM council I decided to make something similar, but leverage my already existing LLM subscriptions + models running locally via Ollama. I also wanted multi-turn iteration to create divergent then convergent topic analysis before finalisation. #chatgpt #ollama #gemini
110 1
AJ Fisher @ajfisher.social · 08/12/2025
An LLM advisory group using your existing subscriptions and any local models you've already downloaded. Ask, "How to poach an egg at home?" and let them debate an answer. www.youtube.com/watch?v=7Xft... #ai #llm #advisors
youtube.com
LLM Advisors screenshare
YouTube video by Andrew Fisher
100 1
Reposted by AJ Fisher
John Allsopp @johnfallsopp.com · 25/11/2025
via @ajfisher.social I recently learned about three-sided football, and it might sound like a trivial change to the game of football. Turns out it has really interesting implications. en.wikipedia.org/wiki/Three-s...
en.wikipedia.org
Three-sided football - Wikipedia
021 2
AJ Fisher @ajfisher.social · 21/11/2025
Discussing a post I'm writing and exploring critiques of the narrative with Gemini 3. Um, okay.... (Though maybe it's referring to Eric Schmidt?? 😅) #ai #gemini #llms
An image of text provided by the Gemini LLM. Underlined is a phrase "my 80 year old father" written from the perspective of the chat bot
020 2
AJ Fisher @ajfisher.social · 18/11/2025
At the Web Directions Dev Summit the next few days in Sydney. Good kick off and summary of the state of the web and react being a local maximum that has led is astray by @johnfallsopp.com #web #development #wdds
011 1
AJ Fisher @ajfisher.social · 05/11/2025
The last week has been lots of strategy & innovation workshops with clients. AI played a role in supporting these activities but deep, creative work is a place where humans really shine and AI should support that. #ai #business #strategy #innovation ajfisher.me/2025/11/05/s...
ajfisher.me
Strategy is still a human activity
A week of deep strategy work with clients highlights the uniquely human capabilities that we need to keep as businesses transition to using more AI within the enterprise and across their workforces.
020 2
AJ Fisher @ajfisher.social · 29/10/2025
Uber is better in a lot of ways for many types of trips but mass dispersed transit out of an airport with no train line isn't it. Simplicity is key.
000 0
AJ Fisher @ajfisher.social · 29/10/2025
I've done the taxi ride a hundred times, maybe more over decades. Yes, Melbourne taxis suck and need modernisation but a taxi rank is a simple and optimised process. Contrast getting off in Sydney the other day, walked out at peak time & got a taxi in 2-3 mins. Perfectly fine.
110 1