Current status — working on my talk for #craftconf looking out over the Danube.
May 26, 2025 at 1:27 PM
Current status — working on my talk for #craftconf looking out over the Danube.
Conference season is heating up: see you there
- AI Engineer Summit www.ai.engineer/summit/2025
- Future of Software www.thefutureofsoftware.com
- Monki Grass monkigras.com
- QCon London qconlondon.com
- CraftConf craft-conf.com/2025
- PlatformCon platformcon.com
Any other Cfps to submit to ?
- AI Engineer Summit www.ai.engineer/summit/2025
- Future of Software www.thefutureofsoftware.com
- Monki Grass monkigras.com
- QCon London qconlondon.com
- CraftConf craft-conf.com/2025
- PlatformCon platformcon.com
Any other Cfps to submit to ?
February 1, 2025 at 11:06 AM
Conference season is heating up: see you there
- AI Engineer Summit www.ai.engineer/summit/2025
- Future of Software www.thefutureofsoftware.com
- Monki Grass monkigras.com
- QCon London qconlondon.com
- CraftConf craft-conf.com/2025
- PlatformCon platformcon.com
Any other Cfps to submit to ?
- AI Engineer Summit www.ai.engineer/summit/2025
- Future of Software www.thefutureofsoftware.com
- Monki Grass monkigras.com
- QCon London qconlondon.com
- CraftConf craft-conf.com/2025
- PlatformCon platformcon.com
Any other Cfps to submit to ?
Thrilled to be presenting at #CraftConf, a conference proving entirely worthy of the FOMO that I have had for it over the years! I'll be presenting on ethical dilemmas in software engineering at 15:50 CEST on Platform 2 -- and it will be livestreamed! https://craft-conf.com/
Craft - Budapest, May 29-30, 2025
Craft is about software craft, which tools, methods, practices should be part of the toolbox of a modern developer and company, and it is a compass on new technologies, trends. You can learn from the best speakers and practitioners in our community.
craft-conf.com
November 3, 2024 at 4:23 AM
Thrilled to be presenting at #CraftConf, a conference proving entirely worthy of the FOMO that I have had for it over the years! I'll be presenting on ethical dilemmas in software engineering at 15:50 CEST on Platform 2 -- and it will be livestreamed! https://craft-conf.com/
Slides for my talk today at #CraftConf on "Andreessen's Corollary: Ethical Dilemmas in Software Engineering", with video to come! https://www.slideshare.net/bcantrill/andreessens-corollary-ethical-dilemmas-in-software-engineering
Andreessen's Corollary: Ethical Dilemmas in Software Engineering
Andreessen's Corollary: Ethical Dilemmas in Software Engineering - Download as a PDF or view online for free
www.slideshare.net
November 3, 2024 at 4:23 AM
Slides for my talk today at #CraftConf on "Andreessen's Corollary: Ethical Dilemmas in Software Engineering", with video to come! https://www.slideshare.net/bcantrill/andreessens-corollary-ethical-dilemmas-in-software-engineering
Video for my #CraftConf talk on "Andreessen's Corollary: Ethical Dilemmas in Software Engineering" is now up: https://www.youtube.com/watch?v=kOURCweZfes
November 3, 2024 at 4:23 AM
Video for my #CraftConf talk on "Andreessen's Corollary: Ethical Dilemmas in Software Engineering" is now up: https://www.youtube.com/watch?v=kOURCweZfes
My #CraftConf talk on ethical dilemmas in software engineering was removed and re-uploaded; for anyone looking for it, it's now here: https://www.youtube.com/watch?v=0wtvQZijPzg
November 3, 2024 at 4:23 AM
My #CraftConf talk on ethical dilemmas in software engineering was removed and re-uploaded; for anyone looking for it, it's now here: https://www.youtube.com/watch?v=0wtvQZijPzg
And here the talk on Who Is Actually Making the Architectural Decisions Right Now? Facilitating Architecture in an AI-Accelerated World:
buff.ly/tMiAp2V
#CollaborativeSoftwareDesign #DomainDrivenDesign #SoftwareArchitecture #SociotechnicalSystems #CraftConf
buff.ly/tMiAp2V
#CollaborativeSoftwareDesign #DomainDrivenDesign #SoftwareArchitecture #SociotechnicalSystems #CraftConf
Who Is Actually Making the Architectural Decisions Right Now? Facilitating Architecture in an AI-Accelerated World - Weave IT
Move from making architectural decisions for teams to making them with teams. Learn how to unlock collaboration and get true buy-in for your software designs. Discover a new approach to technical…
weave-it.org
June 8, 2026 at 7:12 AM
And here the talk on Who Is Actually Making the Architectural Decisions Right Now? Facilitating Architecture in an AI-Accelerated World:
buff.ly/tMiAp2V
#CollaborativeSoftwareDesign #DomainDrivenDesign #SoftwareArchitecture #SociotechnicalSystems #CraftConf
buff.ly/tMiAp2V
#CollaborativeSoftwareDesign #DomainDrivenDesign #SoftwareArchitecture #SociotechnicalSystems #CraftConf
So - erster Tag der CraftConf 2026 startet in Kürze. Bin schon gespannt :)
June 4, 2026 at 5:10 AM
So - erster Tag der CraftConf 2026 startet in Kürze. Bin schon gespannt :)
The Generalize Step In TDD
At CraftConf this year I debuted a talk on Canon TDD.
open.substack.com
June 23, 2025 at 1:19 PM
Quantifying "vigilance toil": troughput × amount to protect × cost to protect. Since there's so much AtP, even very low CtP leads to high VT. So we instead want zero CtP.
Definition of VT: defects you have to worry about.
App builders just needs few enough defects to "work enough"
#CraftConf
Definition of VT: defects you have to worry about.
App builders just needs few enough defects to "work enough"
#CraftConf
June 5, 2026 at 12:20 PM
Quantifying "vigilance toil": troughput × amount to protect × cost to protect. Since there's so much AtP, even very low CtP leads to high VT. So we instead want zero CtP.
Definition of VT: defects you have to worry about.
App builders just needs few enough defects to "work enough"
#CraftConf
Definition of VT: defects you have to worry about.
App builders just needs few enough defects to "work enough"
#CraftConf
Ducked into Arlo Belshee's talk about making environments where AI agents can't fuck up. Screen too small to read from my seat, so I just got what he's saying.
When I got in, he was talking about not letting agents touch his clean git history.
"Agents are not stupid, they're DRUNK"
#CraftConf
When I got in, he was talking about not letting agents touch his clean git history.
"Agents are not stupid, they're DRUNK"
#CraftConf
June 5, 2026 at 12:01 PM
Ducked into Arlo Belshee's talk about making environments where AI agents can't fuck up. Screen too small to read from my seat, so I just got what he's saying.
When I got in, he was talking about not letting agents touch his clean git history.
"Agents are not stupid, they're DRUNK"
#CraftConf
When I got in, he was talking about not letting agents touch his clean git history.
"Agents are not stupid, they're DRUNK"
#CraftConf
-
One of the #CraftConf sponsored talks caught my eye: "Monte Carlo for SaaS" by Zoltan David (ConfigCat).
Weather forecasts work (in part) by markov chains: state machines where the probability of the next state only depends only on the current state. Similar works for Saas
One of the #CraftConf sponsored talks caught my eye: "Monte Carlo for SaaS" by Zoltan David (ConfigCat).
Weather forecasts work (in part) by markov chains: state machines where the probability of the next state only depends only on the current state. Similar works for Saas
June 5, 2026 at 11:52 AM
-
One of the #CraftConf sponsored talks caught my eye: "Monte Carlo for SaaS" by Zoltan David (ConfigCat).
Weather forecasts work (in part) by markov chains: state machines where the probability of the next state only depends only on the current state. Similar works for Saas
One of the #CraftConf sponsored talks caught my eye: "Monte Carlo for SaaS" by Zoltan David (ConfigCat).
Weather forecasts work (in part) by markov chains: state machines where the probability of the next state only depends only on the current state. Similar works for Saas
[this seems like you gotta rewrite your code to match the model, too]
Both styles think workflows should be code, progress durable, long-running work normal. They differ in who does the compute: central orchestrator calling workers vs workers coordinating through database
#CraftConf
Both styles think workflows should be code, progress durable, long-running work normal. They differ in who does the compute: central orchestrator calling workers vs workers coordinating through database
#CraftConf
June 5, 2026 at 10:01 AM
[this seems like you gotta rewrite your code to match the model, too]
Both styles think workflows should be code, progress durable, long-running work normal. They differ in who does the compute: central orchestrator calling workers vs workers coordinating through database
#CraftConf
Both styles think workflows should be code, progress durable, long-running work normal. They differ in who does the compute: central orchestrator calling workers vs workers coordinating through database
#CraftConf
We no longer need separate recovery systems, as failure recovery is a runtime property. #CraftConf
What about nondeterminism? "A non-deterministic system is deterministic in the past — me [the speaker]". once you get the nondet result, you can just redo the same result on replay
What about nondeterminism? "A non-deterministic system is deterministic in the past — me [the speaker]". once you get the nondet result, you can just redo the same result on replay
June 5, 2026 at 9:46 AM
We no longer need separate recovery systems, as failure recovery is a runtime property. #CraftConf
What about nondeterminism? "A non-deterministic system is deterministic in the past — me [the speaker]". once you get the nondet result, you can just redo the same result on replay
What about nondeterminism? "A non-deterministic system is deterministic in the past — me [the speaker]". once you get the nondet result, you can just redo the same result on replay
-
"How durable computing is reshaping distributed systems", Jeremy Edberg, #CraftConf
"Architecture is fossilized constraint". Mainframe → Client-Server → "Stateless". Moving state around is a costly part of distributed systems. We have tons of reliability middleware that doesn't map to business
"How durable computing is reshaping distributed systems", Jeremy Edberg, #CraftConf
"Architecture is fossilized constraint". Mainframe → Client-Server → "Stateless". Moving state around is a costly part of distributed systems. We have tons of reliability middleware that doesn't map to business
June 5, 2026 at 9:38 AM
-
"How durable computing is reshaping distributed systems", Jeremy Edberg, #CraftConf
"Architecture is fossilized constraint". Mainframe → Client-Server → "Stateless". Moving state around is a costly part of distributed systems. We have tons of reliability middleware that doesn't map to business
"How durable computing is reshaping distributed systems", Jeremy Edberg, #CraftConf
"Architecture is fossilized constraint". Mainframe → Client-Server → "Stateless". Moving state around is a costly part of distributed systems. We have tons of reliability middleware that doesn't map to business
The big weakenss of AI CODERS is deskilling. You have to always try understanding before you ask an AI, and compare your understanding to it, and look for weaknesses. AI tempts you to not put in the work, you gotta put in the work
#CraftConf
#CraftConf
June 5, 2026 at 9:12 AM
The big weakenss of AI CODERS is deskilling. You have to always try understanding before you ask an AI, and compare your understanding to it, and look for weaknesses. AI tempts you to not put in the work, you gotta put in the work
#CraftConf
#CraftConf
One simple experiment on a codebase:
1. Delete all code
2. Give the tests to an AI
3. Tell it to write code that passes all the tests
Good way to understand an AI's capability.
What about AI-written unit tests? Those tests add review debt. Can be useful as a refactoring backstop
#CraftConf
1. Delete all code
2. Give the tests to an AI
3. Tell it to write code that passes all the tests
Good way to understand an AI's capability.
What about AI-written unit tests? Those tests add review debt. Can be useful as a refactoring backstop
#CraftConf
June 5, 2026 at 9:02 AM
One simple experiment on a codebase:
1. Delete all code
2. Give the tests to an AI
3. Tell it to write code that passes all the tests
Good way to understand an AI's capability.
What about AI-written unit tests? Those tests add review debt. Can be useful as a refactoring backstop
#CraftConf
1. Delete all code
2. Give the tests to an AI
3. Tell it to write code that passes all the tests
Good way to understand an AI's capability.
What about AI-written unit tests? Those tests add review debt. Can be useful as a refactoring backstop
#CraftConf
The fear: AI can help us understand systems, but what happens when we build systems that are beyond *AI's* capacity to understand?
In any case, before we hit that point, some things we can use AI for understanding. "Explain this to me" is actually helpful, modulo (rarer) hallucinations #CraftConf
In any case, before we hit that point, some things we can use AI for understanding. "Explain this to me" is actually helpful, modulo (rarer) hallucinations #CraftConf
June 5, 2026 at 8:47 AM
The fear: AI can help us understand systems, but what happens when we build systems that are beyond *AI's* capacity to understand?
In any case, before we hit that point, some things we can use AI for understanding. "Explain this to me" is actually helpful, modulo (rarer) hallucinations #CraftConf
In any case, before we hit that point, some things we can use AI for understanding. "Explain this to me" is actually helpful, modulo (rarer) hallucinations #CraftConf
-
"Working effectively with AI code", Michael Feathers, #CraftConf
The main problem with AI software is the same as human software: understandability. Bugs come from us not understanding what the code is doing.
All changes are perilous. But the importance of safety depends on the context
"Working effectively with AI code", Michael Feathers, #CraftConf
The main problem with AI software is the same as human software: understandability. Bugs come from us not understanding what the code is doing.
All changes are perilous. But the importance of safety depends on the context
June 5, 2026 at 8:34 AM
-
"Working effectively with AI code", Michael Feathers, #CraftConf
The main problem with AI software is the same as human software: understandability. Bugs come from us not understanding what the code is doing.
All changes are perilous. But the importance of safety depends on the context
"Working effectively with AI code", Michael Feathers, #CraftConf
The main problem with AI software is the same as human software: understandability. Bugs come from us not understanding what the code is doing.
All changes are perilous. But the importance of safety depends on the context
It can be hard to do this because it's not how managers want us to work.
One troubling trend: "single-player mode", where teams shrink to one programmer. This loses interaction between humans, which is critical to engineering.
#CraftConf
One troubling trend: "single-player mode", where teams shrink to one programmer. This loses interaction between humans, which is critical to engineering.
#CraftConf
June 5, 2026 at 7:39 AM
It can be hard to do this because it's not how managers want us to work.
One troubling trend: "single-player mode", where teams shrink to one programmer. This loses interaction between humans, which is critical to engineering.
#CraftConf
One troubling trend: "single-player mode", where teams shrink to one programmer. This loses interaction between humans, which is critical to engineering.
#CraftConf
@kentbeck.com now handdrawing a diagram: x-axis is features, y-axis futures. At the start, you have maximum futures (/ optionality). Every feature you add burns futures. LLMs burn future at an incredible rate
Used to take years to lose all optionality. LLMs can burn it all in days #CraftConf
Used to take years to lose all optionality. LLMs can burn it all in days #CraftConf
June 5, 2026 at 7:28 AM
@kentbeck.com now handdrawing a diagram: x-axis is features, y-axis futures. At the start, you have maximum futures (/ optionality). Every feature you add burns futures. LLMs burn future at an incredible rate
Used to take years to lose all optionality. LLMs can burn it all in days #CraftConf
Used to take years to lose all optionality. LLMs can burn it all in days #CraftConf
Beck has this model of the "desert and the forest": the desert is working without tests, no agile, poor relationships, where resources are incredibly scarce and you have to do everything perfectly. It's poorly executed, but it still is very profitable software.
#CraftConf
#CraftConf
June 5, 2026 at 7:24 AM
Beck has this model of the "desert and the forest": the desert is working without tests, no agile, poor relationships, where resources are incredibly scarce and you have to do everything perfectly. It's poorly executed, but it still is very profitable software.
#CraftConf
#CraftConf
Day two of #CraftConf! Gonna try skeeting from Edge in hopes it doesn't kill my battery as much as Firefox. I just got in a couple minutes late for @kentbeck.com 's "Forest & Desert & Genie". About how AI unlocks so much creativity but also has so many challenges
June 5, 2026 at 7:18 AM
Day two of #CraftConf! Gonna try skeeting from Edge in hopes it doesn't kill my battery as much as Firefox. I just got in a couple minutes late for @kentbeck.com 's "Forest & Desert & Genie". About how AI unlocks so much creativity but also has so many challenges