complexity.fm Podcast

📄 Show Notes

Anthropic just moved its own code off Git. So is GitKraken in trouble?

Bas Steins and Michal Martinka talk with Chris Griffing, Developer Advocate at GitKraken, Twitch streamer, and serial side-project builder. Bas and Chris met at Commit Your Code in Dallas, where Chris gave a talk on MCP. The conversation goes from protocol details to industry economics.

First up: is MCP dead? Chris says no. With the 2026-07-28 spec, MCP went stateless, so you can now scale servers like Lambdas or Cloudflare Workers instead of babysitting sticky sessions. He also argues that most MCP servers are bad because they just wrap someone's API. Good ones turn several API calls into a single task the agent can finish. We also cover how MCP compares to skills (hint: auth) and why WebMCP might be the easiest way for any web app to become agent-ready. Your browser session becomes the auth layer, and your client-side JavaScript becomes the tool list.

Then it gets spicy. Chris doesn't buy the "SaaSpocalypse": a $50/month SaaS is not worth six hours of a $200k engineer plus tokens plus maintenance forever. Michal pushes back. Companies aren't rational, internal politics reward building new things, and someone once spent two hours arguing over a $1/month EC2 instance. We also ask whether AI is really behind the layoffs or just a convenient scapegoat for post-COVID cuts and offshoring, and why "human slop" was here long before AI slop.

Finally, the elephant in the repo: Anthropic's Scion. Will the Git binary disappear while Git the protocol lives on? Do agents even need branches? Chris's answer: throwing away branches sounds great if you sell tokens, because the merge conflicts burn a lot of them. Plus Kepler, GitKraken's ADE built on ACP, why building your own harness against Claude is a losing game, and whether GitKraken will someday need a new name.

Guest: Chris Griffing (cmgriffing everywhere)

Hosts: Bas Steins and Michal Martinka

⏱️ Topics Covered

00:00 Intro: Chris Griffing and Commit Your Code

00:22 Are MCP servers dead?

01:47 MCP vs. skills: why auth changes everything

02:40 GitKraken as the control plane for GitHub, GitLab, Bitbucket, Jira and more

05:55 Commit Composer: cutting AI-sized commits into readable chunks

07:29 Stateless MCP: JWT-style scaling for agent servers

09:45 WebMCP: making web apps machine-readable for agents

12:11 Context rot is real, whatever the marketing says

14:08 Vercel's agent-browser and WebMCP in practice

17:22 The OpenAI WebMCP Challenge and Chris's side projects

20:10 The SaaSpocalypse myth and the math of build vs. buy

22:36 Mythical Man-Month, stakeholders, and why micro-SaaS exists

24:59 Is AI the scapegoat for layoffs?

26:26 Prompting plus testing: are developers becoming underpaid QA?

29:25 Human slop: localStorage as state management

32:28 Promotion-driven development and Killed by Google

34:03 Don't replace PeopleSoft, build a facade

36:14 Would you vibe-code your own Git GUI?

38:33 Kepler: GitKraken's agentic development environment (ADE)

40:04 ADEs: Conductor, T3 Code and the life after IDEs

42:39 ACP vs. MCP: the two halves of agent tooling

44:14 Anthropic's Scion and Git at agent scale

47:53 Jujutsu, Git compatibility, and Git as a protocol

50:20 Angular, Google's monorepo and $10K of CI per PR

56:25 Do agents need branches at all? Waiting for Git 2

59:25 From Axosoft to GitKraken: will the name survive?

01:00:16 GitLens for solo devs vs. GitKraken for teams

01:02:59 Where to find Chris

🌐 Resources & Links

Chris Griffing

Twitch: https://www.twitch.tv/cmgriffing

X: https://x.com/cmgriffing

GitHub: https://github.com/cmgriffing

LinkedIn: https://www.linkedin.com/in/cmgriffing

GitKraken

Kepler: https://gitkraken.com/kepler

GitKraken MCP Server: https://github.com/gitkraken/mcp/

GitLens for VS Code: https://marketplace.visualstudio.com/items?itemName=eamodio.gitlens

MCP and WebMCP

MCP 2026-07-28 specification announcement: https://blog.modelcontextprotocol.io/posts/2026-07-28/

MCP specification: https://modelcontextprotocol.io/specification/2026-07-28/basic

WebMCP spec draft: https://webmachinelearning.github.io/webmcp

WebMCP on Chrome for Developers: https://developer.chrome.com/docs/ai/webmcp

OpenAI WebMCP Challenge: https://openai.com/webmcp-challenge

Agent Client Protocol (ACP): https://agentclientprotocol.com/get-started/introduction

Agent tooling

Vercel agent-browser: https://github.com/vercel-labs/agent-browser

Conductor: https://www.conductor.build/docs/faq

T3 Code: https://t3.codes/

Version control beyond Git

Redesigning source control at Anthropic (Scion): https://www.anthropic.com/research/redesigning-source-control

Cloudflare Artifacts: https://www.cloudflare.com/products/artifacts/

Jujutsu (jj): https://github.com/jj-vcs/jj

Chris's side projects

Giffium: https://github.com/cmgriffing

Messijo: https://messijo.com/solutions/keyword-monitoring/

Shiki Magic Move: https://github.com/shikijs/shiki-magic-move

Also mentioned

Commit Your Code conference: https://www.commityourcode.com/2026/agenda

Killed by Google: https://killedbygoogle.com/

complexity.fm with Bas Steins and Michal Martinka

📄 Show Notes

LocalStack shut down its free Community edition and put the AWS emulator behind an account and an auth token. One week before that, Hector Ventura open-sourced Floci, a free, MIT-licensed local cloud emulator. It now covers 125 AWS services plus Azure, GCP and Oracle Cloud, and has almost 27,000 GitHub stars.

Bas Steins and Michal Martinka talk with Hector about how he rebuilt AWS from intercepted SDK traffic, why Google was the easiest cloud to emulate and AWS the hardest, and why AI coding agents need a cloud they can break. They also get into why Floci is written in Java instead of Go or Rust, and the less fun side of going viral in open source: PRs from accounts created the same day, CI runners that can't keep up, and no business model yet.

⏱️ Topics Covered

- What Floci is and why local cloud emulation matters

- LocalStack's paywall and Floci's timing

- Rebuilding AWS from intercepted SDK requests, with help from AI

- How close is "close enough" to real AWS? Using real Postgres for RDS and DuckDB for Athena

- Google was easy, AWS was hard, and working with the AWS Smithy team

- Emulators for students, CI/CD and AI agents (an EKS cluster in under a minute instead of 15)

- Reviewing AI-written PRs and keeping the code readable

- Limiting PRs per contributor in the age of "Sloptoberfest"

- Why Java: Quarkus, GraalVM native images, Vert.x and Netty

- From one Reddit post to a community-driven project

- Open source burnout and the true cost of "free"

- Monetization, investor calls and a possible hosted version

- Could Floci become a test layer for Stelvio?

🌐 Resources & Links

Floci on GitHub: https://github.com/floci-io/floci

Floci website: https://floci.io/

LocalStack pricing changes: https://blog.localstack.cloud/2026-upcoming-pricing-changes/

Smithy: https://smithy.io/

Quarkus: https://quarkus.io/

GraalVM: https://www.graalvm.org/

Stelvio: https://stelvio.dev/

Stelvio on GitHub: https://github.com/stelviodev/stelvio

📄 Show Notes

In this episode of complexity.fm, Bas Steins and Michal Martinka sit down with Brandon Bayer, founder of Flightcontrol and creator of Blitz.js, to talk about the messy reality of deploying to AWS without a DevOps team. Brandon walks through why he built Flightcontrol as a Heroku alternative that deploys directly into your own AWS account, why the product ended up being "too limiting" after years of growth, and why he and his co-founder decided to rebuild everything from scratch with a new product called Ravion. The conversation dives deep into platform engineering, Terraform vs OpenTofu, the "DevOps is a mindset, not a role" debate, and how AI agents are reshaping the relationship between developers, DevOps engineers, and infrastructure itself. Brandon also shares the wild story of applying to YC as a "Hail Mary" with zero expectations, getting accepted the same day, and raising a million dollars within a week.

⏱️ Topics Covered

00:00 Introduction to Brandon Bayer and Flightcontrol

01:56 What Flightcontrol does and why it deploys into your own AWS account

02:30 The Heroku free tier shutdown and its (non) impact on Flightcontrol

05:16 Flightcontrol's target customer: teams without a dedicated DevOps engineer

07:11 The typical company journey from PaaS to AWS to platform engineering

09:09 Tech stack behind Flightcontrol and the pain of Next.js app router

10:48 Forking Next.js and building a Rails-like framework, Blitz.js

11:00 Flightcontrol's YC journey: fundraising struggles before acceptance

12:12 The YC interview process and mock interviews with Stripe Atlas

15:34 Getting the YC acceptance call and raising 1 million dollars in a week

18:52 Being in YC's last fully remote COVID batch

19:16 Getting first customers through Twitter and organic growth

24:14 Realizing Flightcontrol was "too limiting" after years of growth

25:16 Introducing Ravion: rebuilding the product from scratch

28:52 How Ravion uses Terraform instead of custom orchestration

30:52 Terraform vs OpenTofu and the licensing drama

35:49 Bringing your own Terraform and custom modules in Ravion

41:14 Cost visibility, logs, and metrics in one unified UI

44:37 Language and cloud agnosticism, RailPack, and Docker generation

48:41 Is DevOps a mindset, not a role? Shift-left and self-service

50:02 AI agents, guardrails, and sandboxed preview environments

55:59 Comparing Ravion to Crossplane and other IDP tools

59:19 Why deterministic infrastructure still matters in the AI agent era

1:03:53 How AI is changing team dynamics between developers and DevOps

1:09:10 Why small teams may be disrupting big companies faster thanks to AI

🌐 Resources & Links

Flightcontrol: https://www.flightcontrol.dev

Flightcontrol Docs: https://www.flightcontrol.dev/docs

Ravion Docs: https://docs.ravion.com

Blitz.js: https://blitzjs.com

Brandon Bayer on X: https://x.com/flybayer

Brandon Bayer on LinkedIn: https://www.linkedin.com/in/brandonbayer1

Flightcontrol on GitHub: https://github.com/flightcontrolhq

Flightcontrol on Y Combinator: https://www.ycombinator.com/companies/flightcontrol

📄 Show Notes

In this episode of complexity.fm, Bas Steins and Michal Martinka sit down with Dominik Sumer, the solo developer behind Snapify and Vimetric, to talk about a decade of building products in public and what happens when that playbook stops working. Dominik walks through how Snapify started as a Stack Overflow clone that flopped, then pivoted into a viral code-snippet-to-video editor built almost entirely through building-in-public momentum on old Twitter. He explains why that same strategy feels broken today, how AI-generated "one shot this" content has flooded social feeds and killed genuine audience engagement, and why he still misses the tight-knit developer communities of 2022 Twitter. The conversation moves into his newer project Vimetric, a privacy-first, open-source web analytics tool, where Dominik unpacks why he open-sourced it, how self-hosting could become the product's next big growth lever, and why he doesn't believe AI coding tools are actually killing SaaS businesses the way social media hot takes claim. The episode closes with a broader debate on whether AI chatbots will eventually replace dashboards entirely, and what that means for indie makers building tools today.

⏱️ Topics Covered

00:00 - Introduction to Dominik Sumer, 12 years as a professional developer

01:20 - The origin story of Snapify, from failed Stack Overflow clone to viral code-snippet tool

04:06 - Why Dominik stepped back from Snapify to start something new

04:29 - Introducing Vimetric, a privacy-first web and product analytics tool

08:00 - Building products from personal need as a solo developer

08:57 - Distribution and marketing through building in public

11:11 - Psychological vulnerability of sharing progress and failures publicly

15:08 - How AI has changed audience reactions and the loss of the "wow" factor

16:33 - The shift in social media from genuine engagement to attention-grabbing content

17:58 - Nostalgia for RSS readers and slower, deeper content consumption

19:13 - Why AI-generated "catchy" posts still perform despite feeling inauthentic

20:40 - The search for a modern equivalent of 2022 Twitter

23:07 - Hacker News as one of the last places for brutally honest feedback

24:22 - Riffing on building a template-free, AI-free social network

26:11 - Finally digging into AI directly

26:57 - Does AI actually cause customer churn for indie tools

30:58 - Why Vimetric is open source and what that decision was based on

33:10 - Does open sourcing actually help discovery and marketing

35:11 - Self-hosting adoption numbers and the missing one-click Docker setup

38:43 - Infrastructure details, Coolify on Hetzner, microservices, multi-tenant setup

40:25 - The chickens-and-eggs analogy for why people still pay for convenience

42:36 - Whether AI meaningfully changes the self-hosting versus SaaS calculus

44:15 - The real threat to SaaS, chatbots replacing dashboards entirely

44:40 - Closing thoughts on calorie trackers, workout apps, and the future interface

🌐 Resources & Links

Snapify: https://snappify.com

Vimetric: https://vimetric.com

Dominik Sumer on Twitter/X: https://twitter.com/dominiksumer

complexity.fm podcast: https://complexity.fm

Hosts: Bas Steins and Michal Martinka

Referenced: Plausible Analytics (mentioned as inspiration for open-source analytics tools)

Referenced: Carbon (mentioned as an early code-snippet sharing tool competitor to Snapify)

📄 Show Notes

Python Is Not Enough: Why Pythonistas Are Falling in Love with Rust

In this episode of complexity.fm, hosts Bas Steins and Michal Martinka sit down with Bob Belderbos — Python veteran, educator, and coach with over 20 years of software development experience. Bob shares his journey from Sun Microsystems sysadmin to co-founding PyBites, to now coaching developers one-on-one and in cohorts on Python, agentic AI, and Rust. We dive deep into why the Python community is increasingly drawn to Rust, what AI is really doing to how developers learn, and whether vibe coding is a superpower or a trap.

⏱️ Topics Covered

0:00 - Introduction and Bob's background: from economics to Sun Microsystems to Python

3:30 - What makes Python a great language and where it falls short

8:00 - The Python vs. Ruby vs. Go vs. TypeScript debate

14:00 - Why Python tooling (Ruff, UV, Polars) is being rewritten in Rust

18:00 - Bob's Rust origin story: a PyCon talk that changed everything

24:00 - How Rust makes you a better Pythonista

28:00 - Type safety in Python: MyPy, Ty, and Pyrefly compared

33:00 - Why Rust is a natural fit for agentic AI systems

38:00 - Bob's coaching philosophy: project-based, high-stakes, real-world

44:00 - The payroll SaaS built in 12 weeks from scratch in Django

50:00 - How AI is changing the way developers learn (for better and worse)

56:00 - Vibe coding, attention span, and the danger of rubber-stamping PRs

1:02:00 - "Deslopification": helping developers clean up AI-generated mess

1:08:00 - Junior to senior, vibe coder to engineer: Bob's three coaching verticals

🌐 Resources & Links

complexity.fm podcast

https://complexity.fm

Bob Belderbos - coaching and courses

https://bobbelderbos.com

PyBites - Python training platform

https://pybit.es

Ruff - fast Python linter written in Rust

https://github.com/astral-sh/ruff

UV - Python package manager written in Rust

https://github.com/astral-sh/uv

PyO3 - Rust bindings for Python

https://pyo3.rs

Polars - DataFrame library with a Rust core

https://pola.rs

MyPy - static type checker for Python

https://mypy-lang.org

Ty - fast type checker by Astral

https://github.com/astral-sh/ty

Pyrefly - type checker by Meta

https://pyrefly.org

The Rust Programming Language (book)

https://doc.rust-lang.org/book

HTMX - hypermedia for the modern web

https://htmx.org

Datastar - reactive hypermedia framework

https://data-star.dev

📄 Show Notes

In this first episode of complexity.fm, Bas and Michal sit down with AI consultant and educator Isaac Flath to get real about AI-assisted coding. Is it actually making senior developers faster? Is vibe coding dead on arrival for production systems? And why is memory still the unsolved problem at the heart of every AI product? No hype, no dogma — just three experienced practitioners comparing notes.

Isaac's AI Engineering Community on Skool

https://www.skool.com/ai-eng/about

Isaac Flath's website and blog

http://isaacflath.com

Next-PLAID retrieval benchmark (LightOn AI)

https://lightonai.github.io/next-plaid

Colgrep — contextual code search tool

https://github.com/colgrep (search "Colgrep" on GitHub)