Otavio Piske (@otavio021) header
Otavio Piske (@otavio021) profile picture

SE at @IBM (ex-@RedHat) working on @ApacheCamel and https://nitter.gq/t.co/YD0zbTMo2V | Tweets his own opinions and thoughts, in 🇺🇸 and 🇧🇷, about technology, work, and life.

Brno, Czech Republic
Joined March 2008
Photo by Otavio Piske (@otavio021)Photo by Otavio Piske (@otavio021)Photo by Otavio Piske (@otavio021)Photo by Otavio Piske (@otavio021)Photo by Otavio Piske (@otavio021)Photo by Otavio Piske (@otavio021)Photo by Otavio Piske (@otavio021)Photo by Otavio Piske (@otavio021)Photo by Otavio Piske (@otavio021)Photo by Otavio Piske (@otavio021)
Today we are introducing Wanaku, the 1st open-source #MCP Router! Wanaku aims to provide unified access, routing & resource management capabilities for your organization and your #AI #Agents. You can meet Wanaku at wanaku.ai/
4
1
5
1,124
I think this is pretty reasonable. To me, the part that resonates the most is: our engineering responsibilities and agency have expanded. If you are the type of engineer that needs to be told what to do, the engineering world is going to get a lot rougher from now on.
Replying to @josephfcox
Here is Proton CTO's AI manifesto, that I obtained multiple copies of 404media.co/manifestos-fear-…
1
67
Curitiba, pleasantly warm? Maybe ... Still, the weather there is still awful - even more so in an El Niño year.
Imagine living somewhere that has 12 months of pleasantly warm weather.
78
My friend Roan started a YouTube channel about #AI, and he is having a series of videos about Wanaku. 🎬 In the first episode, we talk about governing AI agents with Wanaku. More videos are on the way. Check it out: youtube.com/watch?v=RxQRJvUZ…
1
1
1
113
The upcoming Wanaku 0.3.0, to be released in a few days, makes it a lot easier to run @ApacheCamel routes in a governed way. Given a running @WanakuAI instance on #OpenShift or #Kubernetes, it's essentially: wanaku cloud run integration my-integration.yaml
1
807
The operator takes care of the rest. The integration gets added as a forward. It's tools automatically registered on startup. Its #MCP tools, resources and prompts become guarded by Wanaku's policies and evaluators.
1
1
38
In one of those life mysteries that is almost impossible to explain, @PayPal decided to replace their perfectly functional native iOS app with some web-based app, ugly and slow as hell … what is wrong with you, folks? 🤦🏽‍♂️🤦🏽‍♂️
3
1
2
181
And it consumes 700+ megabytes versus 488 from the old native one … hell no I am keeping that one 😂
1
33
BREAKING: @_praxis_ai, a high-performance and security-first proxy server and framework, which also serves as the core framework for @WanakuAI, has just been accepted into the @CloudNativeFdn Sandbox 🎉🎉🎉
1
3
6
242
The good news, I think, is that a lot of what helped make good software successful in the past, remains just as important for AI relevance nowadays: good documentation, seamless upgrade paths, well-thought user experience, etc. 👇🏽
The creator of rails (@dhh, if you've been living under a rock) saying 3% of the code he writes/generates is Ruby, and they're rebuilding HEY as native applications with a rust backend should tell you so much about the changes that are already happening. If what you cared about was ergonomics, developer happiness, productivity, and individual creativity (the things he has touted his entire career) - the single best way to get all those things today is to learn how to build (for yourself and for others) with agents. We see this every day in our own work, and with folks using swamp. The distance between "I need something that works like this" and "here it is" is so short. Here's a quote from yesterday on discord: "I still haven't read a single line of any code created by my agent in the swamp ecosystem it's excellent even that is a step-change in my book, it's intent->solution" This works because we put engineering effort into making the harnesses you work with produce high quality code out of the box. They don't have to read a single line of the code because the code works, and when it doesn't, it's trivial for the agent to fix it. Because the patterns are consistent, the boundaries are clear, and the software it produces is well factored. We make the distance between your idea and seeing the outcome as short as possible. The same is true for your own application. The sooner you start building the machine that builds the machine, the faster the loop will be between your creative spark and the outcome you're looking for. It's not giving all your engineering skill over to the agent - it's putting all that skill to work creating the system that allows your creativity to be expressed quickly, and to use that speed to enhance its quality for yourself and your users. A few tips: 1) That the craft of writing code was your job. The craft, for most of us, was in service to the outcomes we wanted to produce. Outcomes were always the job. 2) Which language you choose, and at what level it operates, is now a choice to be made based on the operational and practical needs of the software rather than your taste as a developer. It doesn't matter if you like writing Rust, Go, TypeScript, or Python. What matters is the needs of the application you're building, and the language/runtime's ability to deliver it. 3) Software Architecture (like DDD) matters more now than it has ever mattered. This is how you avoid vibe coded slop that falls over at scale. Learn how to attach the vocabulary of how your program behaves architecturally to the domain it works in to the implementation in whatever the underlying language is. Then you can refactor, assess it's shape, and maintain it without having to understand every line (because you understand the patterns). 4) Don't be afraid to build what works for you. This is especially true for internal applications and personal productivity. You can have a system that works the way you want it to work, and every investment you make in this pays off hugely in the day to day. 5) Let yourself swim in the deep end of the pool. Yes, you have to learn how to work this way. Yes, you will discover that some things you really believed in are no longer relevant. You will also learn new ways of working that will please you greatly. Resist the temptation to be practical everywhere - start someplace where you can experience working this way. That will teach you how to carry it forward creatively into the other places you work. 6) Hang out with other people working this way, every day. The Swamp discord is hopping (pun intended) with these folks. This is a very creative time. You need to be hanging out and sharing with others all the time.
1
3
161
The bad news is that switching languages, frameworks, libraries, etc is much easier in this new reality. Now, there's a higher risk of your users flocking away if they can't get those agents to do what they need from your software.
1
37
Everyone's talking about Jev from @typesafeai and it made me eager to try on Wanaku. So, I implemented an evaluator for Wanaku that uses it to classify whether an operation is safe to execute or not. asciinema.org/a/bdJJ06GYkBoz…
1
1
83
This brief screencast to show it in action evaluating an MCP tool call request for a production system and a development one. It's all model-based. There's no if "prod" else here.
1
26
Just joined @typesafeai's waitlist ... I'm very curious to see how Wanaku's evaluators work with it 👀
1
66
If my understanding is correctly, I think it can fit very nicely into the decision engine blocking/approving tool decisions in wanaku.ai.
29
Interested in creating a GitHub coding bot? I described the steps that I followed in order to create one using Claude Code and Apitomy Axiom . Check it out at: orpiske.net/posts/building-a…
1
100
1/4 Wanaku is evolving — and our MCP routing ecosystem now uses both Rust and Java. The core routing engine is moving to #Rust, while the Camel Integration Capability, CLI, and Kubernetes Operator remain Java/Quarkus. Why? Different problems, different strengths. 🧵
1
1
1
163
2/4 The Rust side is becoming our LLM / A2A / MCP routing engine, built on Praxis. Its filter pipeline handles namespace isolation, conversation classification, tool execution, response interception — and WASM-based governance, with custom logic written in Rust or JavaScript.
1
1
41
I am trying to get better at writing code in #Rust and I’ve been working on a new side project purely for fun and educational purposes, and I want to to share it with you - hoping it can be useful: Humble Knowledge Builder! It's a CLI tool for automated dataset creation.
1
3
897
The idea is simple: you point hkb at a repository, and it automatically scans for all documentation files. It then hooks up with an LLM to generate a custom dataset composed of synthetic questions and answers based on your docs.
1
27
I'm learning Rust with Codex and, OMG, it's fucking awesome! I wish I had support like that when I started my journey's in C and Java (a long time ago).
1
141