
Mario Zechner wrote of handing the libGDX project to a team of core contributors in 2016 and later worked on the open-source project pi. The items below are things Zechner wrote in blog posts on mariozechner.at about tools, open source, coding agents and charity projects. — “I've sold out,” Mario Zechner (mariozechner.at, April 8, 2026).
## Part 1: The Game Development Journey & libGDX
- On Youthful Side Projects: In the blog post “Two years in review” (July 15, 2024), Zechner wrote: "I relived some of my youth by building a raycaster engine called". — “Two years in review,” Mario Zechner (mariozechner.at, July 15, 2024).
- On Never Commercializing libGDX: In the blog post “I've sold out” (April 8, 2026), Zechner wrote: "I never commercialized libGDX, unless you consider it commercialization to build a proprietary piece of software like Spine on top of it". — “I've sold out,” Mario Zechner (mariozechner.at, April 8, 2026).
- On Idiomatic Ports: In the blog post “Prompts are code, .json/.md files are state” (June 2, 2025), Zechner wrote: "We need idiomatic ports that preserve the same API surface in a way that feels natural for each language". — “Prompts are code, .json/.md files are state,” Mario Zechner (mariozechner.at, June 2, 2025).
- On a Stable API Surface: In the blog post “Prompts are code, .json/.md files are state” (June 2, 2025), Zechner wrote: "What makes this tractable is that we maintain the same API surface across all runtime implementations". — “Prompts are code, .json/.md files are state,” Mario Zechner (mariozechner.at, June 2, 2025).
- On From Blinking LEDs to Circuit Boards: In the blog post “Year in Review 2025” (December 22, 2025), Zechner wrote: "It was quite a journey, going from the basic blinking LED to self-designed printed circuit boards". — “Year in Review 2025,” Mario Zechner (mariozechner.at, December 22, 2025).
- On Opening Up Devices: In the blog post “Year in Review 2025” (December 22, 2025), Zechner wrote: "Anytime something breaks, which happens surprisingly little, the boy and I get to unscrew the device together, look inside its internals, and dig around in them". — “Year in Review 2025,” Mario Zechner (mariozechner.at, December 22, 2025).
- On Work for Non-Nerds: In the blog post “Two years in review” (July 15, 2024), Zechner wrote: "Starting in 2022, my work shifted towards being more accessible to non-nerds". — “Two years in review,” Mario Zechner (mariozechner.at, July 15, 2024).
- On Meaning in the Real World: In the blog post “Two years in review” (July 15, 2024), Zechner wrote: "I blame my now 3-year-old, who somehow manages to replace some of my old man grumpiness with a desire to do things that have actual, if limited, meaning in the real world". — “Two years in review,” Mario Zechner (mariozechner.at, July 15, 2024).
- On Writing the Tool You Need: In the blog post “Two years in review” (July 15, 2024), Zechner wrote: "I needed a bitmap font generator for one of the r96 articles, so I wrote one using web tech". — “Two years in review,” Mario Zechner (mariozechner.at, July 15, 2024).
- On No Regrets: In the blog post “I've sold out” (April 8, 2026), Zechner wrote: "Zero regrets, worked out beautifully, opened many doors, and got me in touch with a lot of very interesting (and kind) people". — “I've sold out,” Mario Zechner (mariozechner.at, April 8, 2026).
## Part 2: The Philosophy of Tools & Architecture
- On Writing the Gestalt by Hand: In the blog post “Thoughts on slowing the fuck down” (March 25, 2026), Zechner wrote: "Anything that defines the gestalt of your system, that is architecture, API, and so on, write it by hand". — “Thoughts on slowing the fuck down,” Mario Zechner (mariozechner.at, March 25, 2026).
- On Simple Predictable Tools: In the blog post “What I learned building an opinionated and minimal coding agent” (November 30, 2025), Zechner wrote: "I'm a simple boy who likes simple, predictable tools". — “What I learned building an opinionated and minimal coding agent,” Mario Zechner (mariozechner.at, November 30, 2025).
- On Context Engineering: In the blog post “What I learned building an opinionated and minimal coding agent” (November 30, 2025), Zechner wrote: "I learned that context engineering is paramount. Exactly controlling what goes into the model's context yields better outputs, especially when it's writing code". — “What I learned building an opinionated and minimal coding agent,” Mario Zechner (mariozechner.at, November 30, 2025).
- On Building on Provider SDKs: In the blog post “What I learned building an opinionated and minimal coding agent” (November 30, 2025), Zechner wrote: "Building on top of the provider SDKs directly gives me full control and lets me design the APIs exactly as I want, with a much smaller surface area". — “What I learned building an opinionated and minimal coding agent,” Mario Zechner (mariozechner.at, November 30, 2025).
- On Inspecting Every Interaction: In the blog post “What I learned building an opinionated and minimal coding agent” (November 30, 2025), Zechner wrote: "I want to inspect every aspect of my interactions with the model". — “What I learned building an opinionated and minimal coding agent,” Mario Zechner (mariozechner.at, November 30, 2025).
- On a Documented Session Format: In the blog post “What I learned building an opinionated and minimal coding agent” (November 30, 2025), Zechner wrote: "I also want a cleanly documented session format I can post-process automatically". — “What I learned building an opinionated and minimal coding agent,” Mario Zechner (mariozechner.at, November 30, 2025).
- On Tokens in Context: In the blog post “Year in Review 2025” (December 22, 2025), Zechner wrote: "What matters more is that whatever tool you use, it doesn't shit a ton of tokens into your context". — “Year in Review 2025,” Mario Zechner (mariozechner.at, December 22, 2025).
- On Commit-Hook Checks: In the blog post “Year in Review 2025” (December 22, 2025), Zechner wrote: "I found setting up forced type checking, linting, and what have you as part of a commit hook to be far more effective at making the agent fix its errors". — “Year in Review 2025,” Mario Zechner (mariozechner.at, December 22, 2025).
- On Agents and the Command Line: In the blog post “MCP vs CLI: Benchmarking Tools for Coding Agents” (August 15, 2025), Zechner wrote: "most often the GitHub MCP Server will lead to much worse results than just letting the agent run the command line tool directly". — “MCP vs CLI: Benchmarking Tools for Coding Agents,” Mario Zechner (mariozechner.at, August 15, 2025).
## Part 3: The RoboVM Era & The Reality of Open Source
- On a Closed-Sourced Core: In the blog post “I've sold out” (April 8, 2026), Zechner wrote: "A short while later Xamarin closed-sourced our open-source RoboVM core, quickly followed by Xamarin selling to Microsoft". — “I've sold out,” Mario Zechner (mariozechner.at, April 8, 2026).
- On a Sale That Sucked: In the blog post “I've sold out” (April 8, 2026), Zechner wrote: "While there was some monetary gain, everything about this fucking sucked". — “I've sold out,” Mario Zechner (mariozechner.at, April 8, 2026).
- On Writing the Bad News: In the blog post “I've sold out” (April 8, 2026), Zechner wrote: "I was also the guy who had to write the “Sorry, no more OSS, suckers” blog post". — “I've sold out,” Mario Zechner (mariozechner.at, April 8, 2026).
- On Abuse Without Control: In the blog post “I've sold out” (April 8, 2026), Zechner wrote: "You cannot imagine what people called me on social media and via email, despite me having zero control over the situation (I was not a majority shareholder)". — “I've sold out,” Mario Zechner (mariozechner.at, April 8, 2026).
- On Souring on VC and OSS: In the blog post “I've sold out” (April 8, 2026), Zechner wrote: "RoboVM going closed-source was devastating, and it made me sour on both VC-funded startups and OSS". — “I've sold out,” Mario Zechner (mariozechner.at, April 8, 2026).
- On the Community Forking: In the blog post “I've sold out” (April 8, 2026), Zechner wrote: "A few days after RoboVM's open-source core was closed-source, a handful of libGDX contributors forked the old RoboVM repo and got to work". — “I've sold out,” Mario Zechner (mariozechner.at, April 8, 2026).
- On a Fork in Days: In the blog post “I've sold out” (April 8, 2026), Zechner wrote: "Within days, libGDX games could be compiled via the fork". — “I've sold out,” Mario Zechner (mariozechner.at, April 8, 2026).
- On Delayed Open Source: In the blog post “I've sold out” (April 8, 2026), Zechner wrote: "Free to use, source available, and they convert to full open-source after a set period via Delayed Open Source Publication (DOSP)". — “I've sold out,” Mario Zechner (mariozechner.at, April 8, 2026).
- On Having Seen Some Things: In the blog post “I've sold out” (April 8, 2026), Zechner wrote: "Because I want you to know that I've seen some shit when it comes to OSS and commercial efforts". — “I've sold out,” Mario Zechner (mariozechner.at, April 8, 2026).
## Part 4: Radical Minimalism & "YOLO" Engineering
- On Messages Baked Into the Weights: In the blog post “Armin is wrong and here's why” (November 22, 2025), Zechner wrote: "But the messages abstraction is baked into the model weights via the prompt format". — “Armin is wrong and here's why,” Mario Zechner (mariozechner.at, November 22, 2025).
- On Leaving Out What You Do Not Need: In the blog post “What I learned building an opinionated and minimal coding agent” (November 30, 2025), Zechner wrote: "My philosophy in all of this was: if I don't need it, it won't be built". — “What I learned building an opinionated and minimal coding agent,” Mario Zechner (mariozechner.at, November 30, 2025).
- On Crimes Against Statelessness: In the blog post “Armin is wrong and here's why” (November 22, 2025), Zechner wrote: "Then providers started committing crimes against statelessness". — “Armin is wrong and here's why,” Mario Zechner (mariozechner.at, November 22, 2025).
- On Retaining Opaque Data: In the blog post “Armin is wrong and here's why” (November 22, 2025), Zechner wrote: "Yes, you now have to retain opaque data sent from the server as part of your client-side state. But that's no different than appending the LLM response". — “Armin is wrong and here's why,” Mario Zechner (mariozechner.at, November 22, 2025).
- On a Stateless Server: In the blog post “Armin is wrong and here's why” (November 22, 2025), Zechner wrote: "As far as you're concerned, the server is still essentially stateless". — “Armin is wrong and here's why,” Mario Zechner (mariozechner.at, November 22, 2025).
- On Whether It Is a Sync Problem: In the blog post “Armin is wrong and here's why” (November 22, 2025), Zechner wrote: "Is this a sync problem? No. The server has the file, you have a reference". — “Armin is wrong and here's why,” Mario Zechner (mariozechner.at, November 22, 2025).
- On an Authoritative Server: In the blog post “Armin is wrong and here's why” (November 22, 2025), Zechner wrote: "Server is authoritative. Standard resource management". — “Armin is wrong and here's why,” Mario Zechner (mariozechner.at, November 22, 2025).
- On Not Claiming Best Solutions: In the blog post “What I learned building an opinionated and minimal coding agent” (November 30, 2025), Zechner wrote: "I'm not claiming my solutions are the best, but they've been working pretty well throughout various agentic and non-agentic LLM projects". — “What I learned building an opinionated and minimal coding agent,” Mario Zechner (mariozechner.at, November 30, 2025).
- On Testing Across Providers: In the blog post “What I learned building an opinionated and minimal coding agent” (November 30, 2025), Zechner wrote: "To ensure all features actually work across the gazillion of providers, pi-ai has a pretty extensive test suite covering image inputs, reasoning traces, tool calling, and other features you'd expect from an LLM API". — “What I learned building an opinionated and minimal coding agent,” Mario Zechner (mariozechner.at, November 30, 2025).
## Part 5: The Rise of Agentic AI & The "Pi" Coding Agent
- On Nobody Knowing Yet: In the blog post “Year in Review 2025” (December 22, 2025), Zechner wrote: "my summary of a year of working with coding agents is: nobody knows yet how to do this properly". — “Year in Review 2025,” Mario Zechner (mariozechner.at, December 22, 2025).
- On Engineering Context: In the blog post “Prompts are code, .json/.md files are state” (June 2, 2025), Zechner wrote: "What we need when using coding agents on bigger codebases is a structured way to engineer context". — “Prompts are code, .json/.md files are state,” Mario Zechner (mariozechner.at, June 2, 2025).
- On Prompts as Code: In the blog post “Prompts are code, .json/.md files are state” (June 2, 2025), Zechner wrote: "we write "code" in the form of prompts that execute on the LLM to produce the actual code that runs on real CPUs". — “Prompts are code, .json/.md files are state,” Mario Zechner (mariozechner.at, June 2, 2025).
- On Context Degradation: In the blog post “Prompts are code, .json/.md files are state” (June 2, 2025), Zechner wrote: "As your session progresses and pulls in more files, tool outputs, and other data, things start falling apart around 100k tokens". — “Prompts are code, .json/.md files are state,” Mario Zechner (mariozechner.at, June 2, 2025).
- On Following Execution Flow: In the blog post “Prompts are code, .json/.md files are state” (June 2, 2025), Zechner wrote: "LLMs still can't follow execution flow all that well". — “Prompts are code, .json/.md files are state,” Mario Zechner (mariozechner.at, June 2, 2025).
- On Resuming With Fresh Context: In the blog post “Prompts are code, .json/.md files are state” (June 2, 2025), Zechner wrote: "You can resume from any point with a fresh context, sidestepping the dreaded compaction issue entirely". — “Prompts are code, .json/.md files are state,” Mario Zechner (mariozechner.at, June 2, 2025).
- On Writing Your Own Harness: In the blog post “Year in Review 2025” (December 22, 2025), Zechner wrote: "I've since abandoned Claude Code and written my own coding agent harness". — “Year in Review 2025,” Mario Zechner (mariozechner.at, December 22, 2025).
- On a System Prompt Cleaning Up: In the blog post “Year in Review 2025” (December 22, 2025), Zechner wrote: "The system prompt told Claude to clean up test files, which explained why it kept deleting things it shouldn't". — “Year in Review 2025,” Mario Zechner (mariozechner.at, December 22, 2025).
- On Context Handoff: In the blog post “What I learned building an opinionated and minimal coding agent” (November 30, 2025), Zechner wrote: "Context handoff between providers was a feature pi-ai was designed for from the start". — “What I learned building an opinionated and minimal coding agent,” Mario Zechner (mariozechner.at, November 30, 2025).
- On Looking Inside the Tool: In the blog post “Year in Review 2025” (December 22, 2025), Zechner wrote: "Around April I started using Claude Code. I got curious about how it actually works". — “Year in Review 2025,” Mario Zechner (mariozechner.at, December 22, 2025).
## Part 6: The Danger of "Slop" & Cognitive Debt
- On Humans as a Bottleneck: In the blog post “Thoughts on slowing the fuck down” (March 25, 2026), Zechner wrote: "A human is a bottleneck". — “Thoughts on slowing the fuck down,” Mario Zechner (mariozechner.at, March 25, 2026).
- On Agents Repeating Errors: In the blog post “Thoughts on slowing the fuck down” (March 25, 2026), Zechner wrote: "An agent has no such learning ability. At least not out of the box. It will continue making the same errors over and over again". — “Thoughts on slowing the fuck down,” Mario Zechner (mariozechner.at, March 25, 2026).
- On Armies of Agents: In the blog post “Thoughts on slowing the fuck down” (March 25, 2026), Zechner wrote: "With an orchestrated army of agents, there is no bottleneck, no human pain". — “Thoughts on slowing the fuck down,” Mario Zechner (mariozechner.at, March 25, 2026).
- On Trusting the Codebase: In the blog post “Thoughts on slowing the fuck down” (March 25, 2026), Zechner wrote: "You realize you can no longer trust the codebase". — “Thoughts on slowing the fuck down,” Mario Zechner (mariozechner.at, March 25, 2026).
- On Limiting Generated Code: In the blog post “Thoughts on slowing the fuck down” (March 25, 2026), Zechner wrote: "Set yourself limits on how much code you let the clanker generate per day, in line with your ability to actually review the code". — “Thoughts on slowing the fuck down,” Mario Zechner (mariozechner.at, March 25, 2026).
- On Local Decisions: In the blog post “Thoughts on slowing the fuck down” (March 25, 2026), Zechner wrote: "Your agents never see each other's runs, never get to see all of your codebase, never get to see all the decisions that were made by you or other agents before they make a change. As such, an agent's decisions are always local". — “Thoughts on slowing the fuck down,” Mario Zechner (mariozechner.at, March 25, 2026).
- On Merchants of Complexity: In the blog post “Thoughts on slowing the fuck down” (March 25, 2026), Zechner wrote: "You let them run free, and they are merchants of complexity". — “Thoughts on slowing the fuck down,” Mario Zechner (mariozechner.at, March 25, 2026).
- On Side Projects: In the blog post “Thoughts on slowing the fuck down” (March 25, 2026), Zechner wrote: "Spending your free time building things is super enjoyable, and most of the time you don't really have to care about code quality and maintainability". — “Thoughts on slowing the fuck down,” Mario Zechner (mariozechner.at, March 25, 2026).
- On Agents Fixing the Mess: In the blog post “Thoughts on slowing the fuck down” (March 25, 2026), Zechner wrote: "So now you hope your agents can fix the mess, refactor it, make it pristine. But your agents can also no longer deal with it". — “Thoughts on slowing the fuck down,” Mario Zechner (mariozechner.at, March 25, 2026).
## Part 7: Hacktivism & Public Good
- On Stress From Manual Work: In the blog post “Two years in review” (July 15, 2024), Zechner wrote: "As the number of requests on Telegram grew, so did her stress levels, something she mentioned in passing in her tweets". — “Two years in review,” Mario Zechner (mariozechner.at, July 15, 2024).
- On a Voucher Website: In the blog post “Two years in review” (July 15, 2024), Zechner wrote: "Built the website, which allowed donors to send us money and refugees to sign up for a voucher". — “Two years in review,” Mario Zechner (mariozechner.at, July 15, 2024).
- On Automating Everything but the Envelopes: In the blog post “Two years in review” (July 15, 2024), Zechner wrote: "Everything except the actual envelope stuffing is automated". — “Two years in review,” Mario Zechner (mariozechner.at, July 15, 2024).
- On Vouchers Sent: In the blog post “Two years in review” (July 15, 2024), Zechner wrote: "We've been able to send out 5,269 vouchers to as many families so far, equaling over €259,000 in donations". — “Two years in review,” Mario Zechner (mariozechner.at, July 15, 2024).
- On Where Donations Come From: In the blog post “Two years in review” (July 15, 2024), Zechner wrote: "While engagement on other social networks like Bluesky or Mastodon is similar, donations are not". — “Two years in review,” Mario Zechner (mariozechner.at, July 15, 2024).
- On Grocery Price Data: In the blog post “Year in Review 2025” (December 22, 2025), Zechner wrote: "I've been advocating for years to get the Austrian government to implement the BWB's (Federal Competition Authority) recommendation from 2023: legalize scraping of grocery price data and require retailers to expose EAN codes". — “Year in Review 2025,” Mario Zechner (mariozechner.at, December 22, 2025).
- On Data Grocers Already Expose: In the blog post “Year in Review 2025” (December 22, 2025), Zechner wrote: "The grocers already expose their data via APIs". — “Year in Review 2025,” Mario Zechner (mariozechner.at, December 22, 2025).
- On Supportive Disagreement: In the blog post “Two years in review” (July 15, 2024), Zechner wrote: "Mr. Schnabel has been very supportive of my efforts, and I think he's a cool dude, even if we do not agree on quite a few topics". — “Two years in review,” Mario Zechner (mariozechner.at, July 15, 2024).
## Part 8: Reflections, Community & The Developer Lifestyle
- On What Mattered: In the blog post “I've sold out” (April 8, 2026), Zechner wrote: "But ultimately, none of this bullshit mattered. What did matter was the community we built around both libGDX and RoboVM". — “I've sold out,” Mario Zechner (mariozechner.at, April 8, 2026).
- On Not Building a Company: In the blog post “I've sold out” (April 8, 2026), Zechner wrote: "I do not want to build my own company around pi". — “I've sold out,” Mario Zechner (mariozechner.at, April 8, 2026).
- On Family First: In the blog post “I've sold out” (April 8, 2026), Zechner wrote: "I want to watch and help him grow up as best as I can. This is, first and foremost, what I want. Everything else is secondary to that". — “I've sold out,” Mario Zechner (mariozechner.at, April 8, 2026).
- On Finding Investors Who Understand: In the blog post “I've sold out” (April 8, 2026), Zechner wrote: "It is extremely hard to find co-founders and investors who understand this space". — “I've sold out,” Mario Zechner (mariozechner.at, April 8, 2026).
- On a Commercial Topping: In the blog post “I've sold out” (April 8, 2026), Zechner wrote: "Make pi OSS sustainable by putting a small team together and adding a commercial topping, that doesn't go against the OSS spirit and the community". — “I've sold out,” Mario Zechner (mariozechner.at, April 8, 2026).
- On No Commercial Value: In the blog post “I've sold out” (April 8, 2026), Zechner wrote: "My European brain thinks pi is just another small, mildly useful OSS project of mine with no commercial value". — “I've sold out,” Mario Zechner (mariozechner.at, April 8, 2026).
- On Conducting Yourself in Heated Discussions: In the blog post “I've sold out” (April 8, 2026), Zechner wrote: "The one thing that differentiated Armin from other internet trolls was the way he conducted himself in these heated discussions. He was never emotional or aggressive". — “I've sold out,” Mario Zechner (mariozechner.at, April 8, 2026).
- On Cordial Disagreement: In the blog post “I've sold out” (April 8, 2026), Zechner wrote: "Our discussions would either end in cordial disagreement, or a newfound common understanding. That's extremely rare on the internet". — “I've sold out,” Mario Zechner (mariozechner.at, April 8, 2026).
- On More in Common: In the blog post “I've sold out” (April 8, 2026), Zechner wrote: "Over coffee, Armin and I found we had more in common than we thought". — “I've sold out,” Mario Zechner (mariozechner.at, April 8, 2026).