chad fowler

59K posts

chad fowler banner
chad fowler

chad fowler

@chadfowler

programmer, general partner @blueyard, author of upcoming book Crypto & Web3: The Good Parts. prev @wunderlist, @microsoft, @rubycentral, @rubyconf @railsconf

New York, USA Katılım Şubat 2007
4.9K Takip Edilen305.7K Takipçiler
chad fowler retweetledi
Daniel Ospina
Daniel Ospina@_Daniel_Ospina·
holy fu**! Started using our newly designed memory and reasoning system for agents, called Tortoise, in production. After every architecture planning or strategy session, I ask the agents to provide feedback on whether Tortoise was useful. Their response when doing an architecture planning and scoping task: Meta-framework finding: The tortoise graph surfaced what linear research didn't — Vector Clocks is the central concern. Causal ordering isn't one pattern among many; it's the architectural constraint that determines viability of everything else. Build causal ordering first; 7 other patterns become near-trivial. We might have a breakthrough in agentic reasoning for complex tasks! Back to testing more, fingers crossed everything checks, and you might be able to try this yourself next week or the one after.
English
1
1
8
618
Sergey Fedorov
Sergey Fedorov@sergefdrv·
@chadfowler Any chance of learning about what you were going to present from elsewhere?
English
1
0
0
11
chad fowler
chad fowler@chadfowler·
In the year 2027 it will be seen as irresponsible for humans to write their own code and insanely inefficient for humans to review all code created by agents Come hear me speak in London on June 1st so you can either listen to why I believe this is true and how we make this okay or at least yell at me in the hall afterward  😅
AI Native Dev@ainativedev

What if the future of software isn’t maintaining systems… but continuously rebuilding them? Chad Fowler ( @chadfowler ) is joining AI Native DevCon London 2026 with one of the most thought-provoking architectural conversations of the event. A technologist, investor, author, and longtime voice in software engineering, Chad has spent decades shaping how developers think about systems and careers. From co-founding Ruby Central to leading engineering organizations at Microsoft and Wunderlist, his work has consistently challenged industry assumptions. Now, through what he calls Phoenix Architecture, he’s asking a much bigger question. What changes when software becomes easier to regenerate than to fully understand? In his session, An Architectural Approach to Regenerative Software, Chad explores how AI is shifting the economics of software development and why many of our existing assumptions about stability, maintenance, and architecture may no longer hold. What this session explores • Why AI changes the balance between maintenance and regeneration • How to think about systems where implementations are constantly replaceable • What parts of software architecture actually need to remain stable • How trust, durability, and coherence evolve in AI-native systems • The human costs of instability and what architects must protect This isn’t a framework talk or a checklist. It’s a deeper exploration of how software engineering itself may need to evolve as agents reshape the mechanics of building systems. If you enjoy talks that challenge foundational assumptions and leave you thinking long after they end, this is one to catch. Join us in London or online: tessl.io/devcon/ (use AIND-X-BB-20 for 20% discount)

English
3
4
12
1.4K
chad fowler retweetledi
Irina Nazarova
Irina Nazarova@inazarova·
I published ruby.evilmartians.com Ruby LLM Discoverability Scorecard after watching @chadfowler 's talk at @RubyConfAT - to ensure that we, the @rubylangorg and @rails community, do our best to be in training data and in agent retrieval. Just updated the data. Here's what's improved in 30 days: - llms.txt +3: @_maintainable , @AppSignal , @flydotio - content negotiation +1 @AppSignal - .md route:+1 @AppSignal - sitemap: +2 @bridgetownrb @any_cable - robots allows AI: +1 Karafka @maciejmensfeld Good job!
English
6
8
45
4.8K
chad fowler
chad fowler@chadfowler·
Thank you for taking the time and energy to read, process, and critique! I have answers on the spec == code thing. They might be wrong but they also aren't just code. Short version: my thinking is that specs are evolving metadata with, among other things, real production runtime output as input. Specs are extracted from natural language (or even code) and then my bias is to store these as nodes in a graph. Specs are never static. We now have the advantage that LLMs can process lots of text, make semantic inferences, and help us fill the gaps of what a full spec really needs, over time. I don't think eg new programming languages for specs are a good idea. They might be a good intermediate representation if plugged into a proper ontology system so you could do semantic type checking etc but I think the quest to create the perfect language for specifications leads exactly to the punch line of that comic.
English
1
2
4
445
chad fowler
chad fowler@chadfowler·
@alediaferia You're right about all of it. Now it's just a matter of completeness and speed.
English
0
0
1
20
Alessandro Diaferia
Alessandro Diaferia@alediaferia·
@chadfowler In a way, it’s already like this, even before AI. Most successful tech orgs treat production as the most important signal to understand where to go next. With AI, most of the phases before shrink significantly, definitely. Also, for some reason I am assuming you are implying AI.
English
1
0
0
53
chad fowler
chad fowler@chadfowler·
For most of software history, production has been the end of the pipeline. Build. Test. Deploy. Run. We’re approaching a world where production becomes the beginning of the next iteration.
English
5
4
19
3.2K
chad fowler retweetledi
Omer Shlomovits
Omer Shlomovits@OmerShlomovits·
Introducing HyperQuant: A new Lattice-based compression algorithm that significantly improves LLM key-value cache memory over TurboQuant and OCTOPUS, and with higher accuracy! But it gets better - we apply the same compression technique to weights, outperforming Higgs! Finally, (for the first time ever?), we show how to compress video DiT weights. demo on LTX-2. Easy to integrate: we release a reference kernel code with the paper, under MIT license. This is a calibration-free, post-training quantization. Disclaimer: performance will get much better in next releases. Links and key results in next tweet.
GIF
English
3
15
54
9.2K
chad fowler
chad fowler@chadfowler·
Every bug report, support ticket, incident, and customer request is a specification. We just haven’t had systems that could consume them directly. aicoding.leaflet.pub/3mjx4erlboc2l
English
0
2
2
786

Keşfet

@thorstenball @sqs @AmpCode @sergefdrv @mipsytipsy @RubyConfAT @rubylangorg @rails