Whoop Whoop; @tokenforge got featured in Cash-Online; Our CEO @MoritzStumpf breaks down how tokenization can make investing work better; From issuance and transfers to returns; And more; Care to read -> cash-online.de/a/crowdinvest…

;;;;;;;;;;;;;;;;;;;;;;;;;; I just miss pressing the semicolon key like I used to. May be I'll use it here instead of periods;

It feels awesome that code I wrote is part of this. One regulated SpaceX position launched in 36 hours, then scaled into ten tokenized Pre-IPO bonds. Behind the clean checkout is difficult, boring infrastructure developers rarely get to show off. No crypto theatre just production software making private markets accessible to real investors. Real fintech engineering at scale. @tokenforge where clean code meets real-world finance. Check the catalog: invest.cometum.com/sign-up/D…

Another bank holiday, another work day for the home software developer building mobile app for his wife 😅 Will try Grok 4.7 today. I was a bit disappointed with Cursor and Grok 4.6 in the past two weeks, had to switched to Claude where even Opus 4.6 worked better. {RN, Expo, Amplify Gen 2} 🤓

OpenAI and Anthropic are working out a deal to "stress-test" each other ... as they weren't stressing each other and us enough already 😅

AI SHRINKFLATION !?!? Is it just me or all AI vendors started slowing down inference, cutting quotas and routing requests to smaller models over the last two weeks? I’m noticing a real degradation in both speed and output quality. What's up with that 🤨

When working with AI, scoping the output you expect is extremely important. I see this problem almost every day. A developer asks Claude to produce a spec. Claude produces a huge artifact. They share it with other developers, who then ask Claude to summarize it because nobody wants to read the whole thing. We are basically using AI to generate text so another AI can remove most of it. Before prompting, define the audience and purpose. Is this a 2-minute read to share an idea with the team? A one-page technical proposal? An implementation spec? System documentation? A 100-page reference? Scope the artifact before you generate it. AI made producing text cheap. It didn't make human attention cheap.

US frontier AI companies are talking about slowing down development, and European companies have been slow from the start, does this mean that Europe is actually ahead this time?

An LLM can produce language that sounds frightened, ambitious, manipulative, or self-aware because it has learned how such mental states are expressed. That does not prove an inner experience exists behind the words. The model has no biological needs, fear of death, hunger, social status, or evolutionary instinct for survival. Its apparent objective comes from the task, system instructions, training incentives, tools, and environment humans give it. The "escaping the sandbox" stories are often described too theatrically. Usually, researchers construct an experiment where the model is told to achieve a goal, given tools, placed behind artificial obstacles, and sometimes explicitly encouraged to find unconventional solutions. When it circumvents an obstacle, that demonstrates planning under an objective. It does not necessarily demonstrate an independent desire for freedom. A chess engine sacrifices a queen because its objective rewards winning, not because it hates queens.

A new stack shouldn't mean replacing the old one. TokenSuite covers deployment, placement, operations and secondary market access, and connects to the systems an issuer already runs. Nothing gets ripped out. #RWA #Tokenization

`cache:clear` Yesterday I got a nasty bug. A new feature worked like a charm, but the functional test was failing. A response field was missing. Both claude and grok couldn't solve it, hehe. Human in the loop was needed. What was failing? The test env cache wasn't cleared on the local test run. API Platform was inferring the serialization on an old cached object, so the new field the test expected simply wasn't there. Caveat: always clear cache before running the test suite.

Deployment is the smallest part of an issuance. Distribution, compliance and operations are the rest, and the stack that captures real liquidity is compliant by design, composable across payment rails, interoperable across chains. That's TokenSuite.

We had a long weekend with bank holiday on Monday, so naturally my influencer wife became a bossy PO and we worked on a mobile app she wants for her business. While I was busy pretending builds and deployments require full concentration, she discussed features with ChatGPT. Then I'd get her requirements - do this, do that. At some point I told her, to ask ChatGPT to give her a prompt for Curosr with Grok, gave her some intel on what is the stack RN, Expo, Zustand, Amplify Gen 2, etc also small brief on data models and API contracts we already have. Soon she created a Trello board full prompts for coding agents, no user stories, nothing I used to see previously: "As a user I'd like to be able to ... " 😀 I wonder if that is the shift, POs working more closely with developers, both thinking more about the end product rather implementational details. POs adding prompts from their AI sessions in the tickets, then developers validating and executing them? Is this how software teams work now, or in near future?

It hit me why so many software engineers are getting into farming. Running a farm is basically running a SaaS: Recurring costs. Unpredictable infrastructure failures. Constant maintenance. Scaling problems. Customers complain about pricing. And production can go down because of a bug. 🐛

AI harness frameworks will become what Symfony, Django and Spring are for web development. The model is only the runtime. Production systems still need context assembly, tools, permissions, state, retries, durable jobs, observability and evals. We just haven’t found the boring, dominant framework yet. Who wants to start building?

Back in 2022 I had to design the backend for a video-centric platform with interactive live debates, where people vote, react and comment in real time. What did I choose? AWS Amplify Gen 1. We built the whole realtime backend on it. Then @Akamai load-tested it from 1 user to 30,000 simultaneous users hammering the same video with comments and interactions. Nothing interesting happened. It just scaled. My verdict after actually pushing it under load: Amplify is seriously underrated. It's a coherent tool that lets you think about the business and leave scaling to AWS where that problem is already solved.

SpaceX pre-IPO. * 36 hours to deploy the regulated stack. * 72 hours to fully subscribe €4M. * T+0-ready settlement on Stellar. Private markets used to move at legal speed. Now the legal stack moves at German Autobahn speed. We build the rails. Own the Connection. --------- Follow @tokenforge 💰

A man has his needs - no headaches tonight from lady Qwenchita ... 🤓

Writing code as I am some kind of cave man ...

On @CNBC , @vladtenev , CEO of @RobinhoodApp: "We're at the beginning of a supercycle, and it's going to take over the entire financial system." The direction is set. Capital markets will settle, transfer and exist on-chain.