Mark Levison
@mlevison.bsky.social
I help repair Bad Scrum and Fake Agile.
I want to help teams be effective, fun places - not feature factories.
I want to help teams be effective, fun places - not feature factories.
Jamie Twiss found newer models silently remove safety checks rather than fail visibly. The result: code that looks right but breaks in production at 2am.
The tool matters less than how you use it.
The tool matters less than how you use it.
February 12, 2026 at 6:50 PM
Jamie Twiss found newer models silently remove safety checks rather than fail visibly. The result: code that looks right but breaks in production at 2am.
The tool matters less than how you use it.
The tool matters less than how you use it.
"Reasoning" models don't reason. They rely on statistical pattern matching—replicating patterns from training data. That's not logical deduction.
More effective than models without it? Sure. But it's still a stochastic parrot.
More effective than models without it? Sure. But it's still a stochastic parrot.
February 11, 2026 at 4:03 PM
"Reasoning" models don't reason. They rely on statistical pattern matching—replicating patterns from training data. That's not logical deduction.
More effective than models without it? Sure. But it's still a stochastic parrot.
More effective than models without it? Sure. But it's still a stochastic parrot.
"Two hours of Sprint Planning. Zero confidence we can deliver."
Sound familiar? The problem isn’t Planning—it’s what happened before you walked in.
Sound familiar? The problem isn’t Planning—it’s what happened before you walked in.
February 11, 2026 at 2:12 PM
"Two hours of Sprint Planning. Zero confidence we can deliver."
Sound familiar? The problem isn’t Planning—it’s what happened before you walked in.
Sound familiar? The problem isn’t Planning—it’s what happened before you walked in.
AI-generated code has 1.7x more issues, 30-41% more tech debt, 39% more cognitive complexity.
Some teams think TDD is the fix. They're solving the wrong problem.
Some teams think TDD is the fix. They're solving the wrong problem.
February 10, 2026 at 7:50 PM
AI-generated code has 1.7x more issues, 30-41% more tech debt, 39% more cognitive complexity.
Some teams think TDD is the fix. They're solving the wrong problem.
Some teams think TDD is the fix. They're solving the wrong problem.
Renaming your teams "Squads" didn't help, did it?
The Spotify Model was a snapshot of one company's culture — autonomy, psychological safety, continuous improvement. Most adoptions skip to the labels and ignore the hard work.
Structure without culture is just a reorg with fancier names.
The Spotify Model was a snapshot of one company's culture — autonomy, psychological safety, continuous improvement. Most adoptions skip to the labels and ignore the hard work.
Structure without culture is just a reorg with fancier names.
February 6, 2026 at 2:12 PM
Renaming your teams "Squads" didn't help, did it?
The Spotify Model was a snapshot of one company's culture — autonomy, psychological safety, continuous improvement. Most adoptions skip to the labels and ignore the hard work.
Structure without culture is just a reorg with fancier names.
The Spotify Model was a snapshot of one company's culture — autonomy, psychological safety, continuous improvement. Most adoptions skip to the labels and ignore the hard work.
Structure without culture is just a reorg with fancier names.
Anthropic claims orgs are building everything with Claude Code at quality. If it works so well, why not share the recipe?
The data elsewhere:
- 1.7x more correctness issues
- 1.64x worse maintainability
- 1.57x more security issues
- Only win? Spelling.
The data elsewhere:
- 1.7x more correctness issues
- 1.64x worse maintainability
- 1.57x more security issues
- Only win? Spelling.
February 5, 2026 at 7:06 PM
Anthropic claims orgs are building everything with Claude Code at quality. If it works so well, why not share the recipe?
The data elsewhere:
- 1.7x more correctness issues
- 1.64x worse maintainability
- 1.57x more security issues
- Only win? Spelling.
The data elsewhere:
- 1.7x more correctness issues
- 1.64x worse maintainability
- 1.57x more security issues
- Only win? Spelling.
AI is making us write more code. That's the problem.
Research on AI-generated code:
→ 1.7x more issues
→ 30-41% more tech debt
→ 39% more complexity
→ Speed gains disappear in months
We're building the wrong thing faster and calling it productivity.
Research on AI-generated code:
→ 1.7x more issues
→ 30-41% more tech debt
→ 39% more complexity
→ Speed gains disappear in months
We're building the wrong thing faster and calling it productivity.
February 4, 2026 at 8:55 PM
AI is making us write more code. That's the problem.
Research on AI-generated code:
→ 1.7x more issues
→ 30-41% more tech debt
→ 39% more complexity
→ Speed gains disappear in months
We're building the wrong thing faster and calling it productivity.
Research on AI-generated code:
→ 1.7x more issues
→ 30-41% more tech debt
→ 39% more complexity
→ Speed gains disappear in months
We're building the wrong thing faster and calling it productivity.
GenAI is making code generation easier, but at what cost?
We're generating more code AND more technical debt. When we 10x code volume but 20x defects, we're not winning—we're compounding problems.
We're generating more code AND more technical debt. When we 10x code volume but 20x defects, we're not winning—we're compounding problems.
February 4, 2026 at 3:35 PM
GenAI is making code generation easier, but at what cost?
We're generating more code AND more technical debt. When we 10x code volume but 20x defects, we're not winning—we're compounding problems.
We're generating more code AND more technical debt. When we 10x code volume but 20x defects, we're not winning—we're compounding problems.
Eight-hour refinement sessions. Incomplete stories. Team burnout.
The obvious fix? Better planning. More process.
The actual root cause? "The system keeps trying to solve a structure problem with a process problem."
The obvious fix? Better planning. More process.
The actual root cause? "The system keeps trying to solve a structure problem with a process problem."
February 4, 2026 at 2:12 PM
Eight-hour refinement sessions. Incomplete stories. Team burnout.
The obvious fix? Better planning. More process.
The actual root cause? "The system keeps trying to solve a structure problem with a process problem."
The obvious fix? Better planning. More process.
The actual root cause? "The system keeps trying to solve a structure problem with a process problem."
Plot twist: Devs using AI feel 20% more productive—but take 19% LONGER to complete tasks. It's a perception gap costing real money.
February 3, 2026 at 7:06 PM
Plot twist: Devs using AI feel 20% more productive—but take 19% LONGER to complete tasks. It's a perception gap costing real money.
Two GenAI code papers, very different quality. 🧵
Cortex: PRs up 20%, incidents up 23.5%, failure rate up 30%. No data shared.
buff.ly/Q40cJFD
Cortex: PRs up 20%, incidents up 23.5%, failure rate up 30%. No data shared.
buff.ly/Q40cJFD
Engineering in the Age of AI: 2026 Benchmark Report - Key Findings | Cortex
Everyone's talking about how AI is transforming software development. We wanted to understand what's actually happening. So we surveyed over 50 engineering leaders and analyzed development metrics…
www.cortex.io
February 3, 2026 at 2:12 PM
Two GenAI code papers, very different quality. 🧵
Cortex: PRs up 20%, incidents up 23.5%, failure rate up 30%. No data shared.
buff.ly/Q40cJFD
Cortex: PRs up 20%, incidents up 23.5%, failure rate up 30%. No data shared.
buff.ly/Q40cJFD
Jobs aren't skill-based, they're task-based. A job is a bundle of tasks.
Travel agents got unbundled into separate booking apps. Stockbrokers into trading platforms + robo-advisors. 🧵
Travel agents got unbundled into separate booking apps. Stockbrokers into trading platforms + robo-advisors. 🧵
cehd.uchicago.edu
February 2, 2026 at 7:06 PM
Jobs aren't skill-based, they're task-based. A job is a bundle of tasks.
Travel agents got unbundled into separate booking apps. Stockbrokers into trading platforms + robo-advisors. 🧵
Travel agents got unbundled into separate booking apps. Stockbrokers into trading platforms + robo-advisors. 🧵
Opposite of #Enshittification recently I’ve replaced smoke alarms and parts of our kitchen faucets. Both First Alert and Delta, get props for products that were easy to install and took far less time than I expected.
February 2, 2026 at 2:28 PM
Opposite of #Enshittification recently I’ve replaced smoke alarms and parts of our kitchen faucets. Both First Alert and Delta, get props for products that were easy to install and took far less time than I expected.
Developers using AI tools feel 20% more productive—but take 19% LONGER to complete tasks. The perception gap is real. 🧵
Where is AI Taking Software Development in 2026?
Explore GenAI's real impact on software development in 2026. Strip away the hype. Focus on what is changing and what the impact will be.
agilepainrelief.com
February 2, 2026 at 2:12 PM
Developers using AI tools feel 20% more productive—but take 19% LONGER to complete tasks. The perception gap is real. 🧵
Systems Thinking is powerful for complex team problems—but hard to apply in the moment. Training makes sense, real dysfunction is tough. 🧵
Systems Thinking with GenAI: Solve Deep Team Problems
GenAI helps you apply Systems Thinking to team problems. Understand the interconnected parts and avoid common traps like local optimization.
agilepainrelief.com
January 30, 2026 at 7:06 PM
Systems Thinking is powerful for complex team problems—but hard to apply in the moment. Training makes sense, real dysfunction is tough. 🧵
KMart used barcodes for faster checkouts. Walmart used them to reimagine their entire business—transforming suppliers into interchangeable parts.
Same tech, wildly different outcomes. 🧵
Same tech, wildly different outcomes. 🧵
January 30, 2026 at 2:12 PM
KMart used barcodes for faster checkouts. Walmart used them to reimagine their entire business—transforming suppliers into interchangeable parts.
Same tech, wildly different outcomes. 🧵
Same tech, wildly different outcomes. 🧵
OReilly has a very good analysis of how AI might play out across all industries in 2026 and beyond: buff.ly/ZRV5iKb
They're not trying make perfect predictions, instead run a range of scenarios and see what happens. Based on the past few weeks, I think that OpenAI is in trouble.
....
They're not trying make perfect predictions, instead run a range of scenarios and see what happens. Based on the past few weeks, I think that OpenAI is in trouble.
....
What If? AI in 2026 and Beyond
The market is betting that AI is an unprecedented technology breakthrough, valuing Sam Altman and Jensen Huang like demigods already astride the world. The
www.oreilly.com
January 29, 2026 at 2:12 PM
OReilly has a very good analysis of how AI might play out across all industries in 2026 and beyond: buff.ly/ZRV5iKb
They're not trying make perfect predictions, instead run a range of scenarios and see what happens. Based on the past few weeks, I think that OpenAI is in trouble.
....
They're not trying make perfect predictions, instead run a range of scenarios and see what happens. Based on the past few weeks, I think that OpenAI is in trouble.
....
I've spent 10+ years encouraging ScrumMasters to use Systems Thinking to understand team challenges. The problem? It's hard to remember all the questions. That's where GenAI actually helps.
I built a Claude skill to save some effort: instead of memorizing a framework, you tell the AI your story.
I built a Claude skill to save some effort: instead of memorizing a framework, you tell the AI your story.
Systems Thinking with GenAI: Solve Deep Team Problems
GenAI helps you apply Systems Thinking to team problems. Understand the interconnected parts and avoid common traps like local optimization.
agilepainrelief.com
January 26, 2026 at 2:12 PM
I've spent 10+ years encouraging ScrumMasters to use Systems Thinking to understand team challenges. The problem? It's hard to remember all the questions. That's where GenAI actually helps.
I built a Claude skill to save some effort: instead of memorizing a framework, you tell the AI your story.
I built a Claude skill to save some effort: instead of memorizing a framework, you tell the AI your story.
Judgement and good taste are even more important in the world of GenAI. Last week, I tried to get #ClaudeCode to help me with A/B testing text in the popups and banner on the Agile Pain Relief website. Ours is a static website, built on Astro and using PostHog for analytics.
Where is AI Taking Software Development in 2026?
Explore GenAI's real impact on software development in 2026. Strip away the hype. Focus on what is changing and what the impact will be.
buff.ly
January 23, 2026 at 4:59 PM
Judgement and good taste are even more important in the world of GenAI. Last week, I tried to get #ClaudeCode to help me with A/B testing text in the popups and banner on the Agile Pain Relief website. Ours is a static website, built on Astro and using PostHog for analytics.
I've argued that while GenAI can reduce critical thinking, using it well actually requires more critical thinking. So how can these tools help develop that skill?
For 10+ years, I've encouraged ScrumMasters to use Systems Thinking for deeper problem understanding.
For 10+ years, I've encouraged ScrumMasters to use Systems Thinking for deeper problem understanding.
January 22, 2026 at 4:25 PM
I've argued that while GenAI can reduce critical thinking, using it well actually requires more critical thinking. So how can these tools help develop that skill?
For 10+ years, I've encouraged ScrumMasters to use Systems Thinking for deeper problem understanding.
For 10+ years, I've encouraged ScrumMasters to use Systems Thinking for deeper problem understanding.
ClaudeCode makes missteps automating tasks. Extracting BufferApp data via Playwright, I asked for LinkedIn→Mastodon→BlueSky scrapers. Got 3 separate codebases (~2/3 duplicated) instead of reusable components. Different CSV formats. Claude ran code in its chat, burning my morning's tokens.
January 21, 2026 at 2:12 PM
ClaudeCode makes missteps automating tasks. Extracting BufferApp data via Playwright, I asked for LinkedIn→Mastodon→BlueSky scrapers. Got 3 separate codebases (~2/3 duplicated) instead of reusable components. Different CSV formats. Claude ran code in its chat, burning my morning's tokens.
Downloading Buffer analytics, everyone said "let Claude drive Playwright." I refused.
Realization: AI automation creates ongoing "cognitive tax"—carefully checking results every run. Instead, have AI write code once. Work harder upfront, but run safely afterward. Failures become self-evident.
Realization: AI automation creates ongoing "cognitive tax"—carefully checking results every run. Instead, have AI write code once. Work harder upfront, but run safely afterward. Failures become self-evident.
January 20, 2026 at 7:12 PM
Downloading Buffer analytics, everyone said "let Claude drive Playwright." I refused.
Realization: AI automation creates ongoing "cognitive tax"—carefully checking results every run. Instead, have AI write code once. Work harder upfront, but run safely afterward. Failures become self-evident.
Realization: AI automation creates ongoing "cognitive tax"—carefully checking results every run. Instead, have AI write code once. Work harder upfront, but run safely afterward. Failures become self-evident.
Agent Psychosis: Are We Going Insane? buff.ly/u0xrCng - I think it should have been called Agents gone wild. It could be a bad movie on Netflix.
Working in small chunks, using BDD/TDD like approaches - I'm getting good results for fewer tokens. ...
Working in small chunks, using BDD/TDD like approaches - I'm getting good results for fewer tokens. ...
Agent Psychosis: Are We Going Insane?
What’s going on with the AI builder community right now?
lucumr.pocoo.org
January 20, 2026 at 2:12 PM
Agent Psychosis: Are We Going Insane? buff.ly/u0xrCng - I think it should have been called Agents gone wild. It could be a bad movie on Netflix.
Working in small chunks, using BDD/TDD like approaches - I'm getting good results for fewer tokens. ...
Working in small chunks, using BDD/TDD like approaches - I'm getting good results for fewer tokens. ...
GenAI is helping to create more code faster. As a Systems Thinker, I ask: what are the unintended consequences of this?
- More work piling up in front of QA
- Increased importance of clear 'specs'
- More technical debt
- Increased security risks
- More work piling up in front of QA
- Increased importance of clear 'specs'
- More technical debt
- Increased security risks
Where is AI Taking Software Development in 2026?
Explore GenAI's real impact on software development in 2026. Strip away the hype. Focus on what is changing and what the impact will be.
agilepainrelief.com
January 16, 2026 at 2:12 PM
GenAI is helping to create more code faster. As a Systems Thinker, I ask: what are the unintended consequences of this?
- More work piling up in front of QA
- Increased importance of clear 'specs'
- More technical debt
- Increased security risks
- More work piling up in front of QA
- Increased importance of clear 'specs'
- More technical debt
- Increased security risks
Asked an LLM for help with a vague user story. It sounded confident, so I went with it. QA found the problem.
LLMs predict tokens—they don't understand context. That confidence is a training artifact, not intelligence. It's a trap. The model can't tell if it should 95% confident or 25%...
LLMs predict tokens—they don't understand context. That confidence is a training artifact, not intelligence. It's a trap. The model can't tell if it should 95% confident or 25%...
January 15, 2026 at 2:12 PM
Asked an LLM for help with a vague user story. It sounded confident, so I went with it. QA found the problem.
LLMs predict tokens—they don't understand context. That confidence is a training artifact, not intelligence. It's a trap. The model can't tell if it should 95% confident or 25%...
LLMs predict tokens—they don't understand context. That confidence is a training artifact, not intelligence. It's a trap. The model can't tell if it should 95% confident or 25%...