iiinova — where your digital self keeps evolving

Raphael De Lio @raphaeldelio.dev · 31/05/2026

The live agent activity stream is possible thanks to NDJSON (Newline Delimited JSON) where each event is sent as an individual JSON object separated by a newline. Allowing the UI to display agent thoughts, tool calls, and intermediate results as they happen without waiting for the workflow to finish

Raphael De Lio @raphaeldelio.dev · 31/05/2026

Experimentations with Spring AI. Multi agent architecture with streaming of what the agent is actually doing while running. #javabubble

261

Raphael De Lio @raphaeldelio.dev · 26/05/2026

AI engineering means designing the context, runtime, and harness around agents so they run reliably, observably, and at scale. In my Redis webinar on May 28, I’ll focus on harness engineering: caching, session management, and rate limiting. RSVP: redis.io/resources/vi...

Raphael De Lio @raphaeldelio.dev · 21/05/2026

Harness engineering is the buzzword of 2026. But the ideas behind it are decades old. Turns out that building scalable agents still requires the same systems engineering skills software engineers have relied on for years to build scalable and sustainable systems. redis.io/resources/vi...

redis.io

Session management, caching, & rate limiting

Solves common real-time application workloads; personalization, search, fraud detection, session management, and AI context.

Raphael De Lio @raphaeldelio.dev · 20/05/2026

OpenAI dropped a tool to verify if an image has been AI generated. And LinkedIn is already using it to tag images that were AI made. The technology is grounded in two mechanisms. One that relies on metadata. Another that relies on imperceptible pixels. Read more: openai.com/index/advanc...

121

Raphael De Lio @raphaeldelio.dev · 19/05/2026

Dizer que não queria ou que esperava uma promoção não eram opções?

Codetaur @vibe-coded.com · 18/05/2026

Reposted by Raphael De Lio

real software is still so much work wtf

71258

Raphael De Lio @raphaeldelio.dev · 16/05/2026

No other Raphael De Lio in the world afaik 😂 raphaeldelio.com

000

Raphael De Lio @raphaeldelio.dev · 14/05/2026

Can you invite me to lobsters?

100

Reposted by Raphael De Lio

Martijn Smit @lostdomain.org · 13/05/2026

yes. just, yes. 🧑‍🍳

011

Raphael De Lio @raphaeldelio.dev · 13/05/2026

Happy to announce that I'm returning as the host of the Dutch Kotlin User Group in Amsterdam after a year and a half hiatus!

030

Raphael De Lio @raphaeldelio.dev · 12/05/2026

I love when users find bugs in my products. It means the product is actually being used.

060

Raphael De Lio @raphaeldelio.dev · 06/05/2026

Only country in western Europe where internet works reliably: the Netherlands

Raphael De Lio @raphaeldelio.dev · 02/05/2026

O vento também consegue fazer essas coisas, não consegue? Mas tenta atirar no vento… não dá em nada

100

Raphael De Lio @raphaeldelio.dev · 02/05/2026

Um corpo fantasmagórico não tem densidade suficiente pra ser afetado por uma bala… A pergunta é mais por que não usam o aspirador de pó já que todo mundo também tem um nos EUA.

media.tenor.com

a video game shows a room with a rocking chair and a game boy

ALT: a video game shows a room with a rocking chair and a game boy

Raphael De Lio @raphaeldelio.dev · 02/05/2026

Bought more shares of Duolingo yesterday. I had previously bought it when learning Dutch a couple years ago and decided to do so again that I’m learning Japanese. Practicing something like one hour a day during a week and I can already read Hiragana. Love this app.

011

I’m addicted to seeing how much energy I’m consuming/producing in realtime. I spend more time on this app than any social media 😂 @zonneplan.bsky.social

9/9 But software engineers should not feel like outsiders in AI. If you know how to make unreliable systems behave reliably in production, you already understand one of the hardest parts of building AI agents. Agentic applications need both systems expertise and model expertise

000

8/9 This does not mean machine learning does not matter. It absolutely does. Machine learning engineers understand model behavior, evaluation, uncertainty, drift, and failure modes. That expertise is critical.

100

7/9 In other words, a lot of agentic AI is still about the same engineering guarantees that have always made distributed systems work. Consistency. Reliability. Security. Observability. Failure handling. Production discipline.

6/9 This is what people are starting to call harness engineering. Harness engineering is the deterministic system you build around the model. The workflows. The validations. The retries. The logs. The tests. The fallbacks. The guardrails.

100

5/9 Building AI agents means understanding: How to handle unreliable outputs How to design safe tool calls How to manage state How to recover from partial failures How to protect data and permissions How to observe behavior in production

4/9 Learning ML fundamentals is valuable. But most software engineers do not need to become machine learning researchers to build agentic applications. Models are available through APIs now. The harder problem is integrating them into real systems.

101

3/9 When I started moving from software engineering into AI engineering, I thought I needed to deeply understand model training, evaluation, feature engineering, research papers, and ML deployment pipelines before I could build useful agents. I was wrong.

100

2/9 In distributed systems, we usually solve this with the Transactional Outbox Pattern. You write the business change and the pending event in the same database transaction. Then another process reliably publishes the event to Kafka.

100

Raphael De Lio @raphaeldelio.dev · 27/04/2026

1/9 If a service commits something to a database and then publishes a message to Kafka, what happens if the database commit succeeds but the Kafka publish fails? This problem has nothing to do with AI. But the mindset behind solving it matters a lot for AI agents.

100

Dude blamed everyone and everything. Blamed Cursor. Blamed the agent. Blamed Railway. Blamed the API. Blamed the token scope. Blamed the lack of confirmation... Assuming the thing had became sentient was just another way of making up for the lack of accountability.

010

Raphael De Lio @raphaeldelio.dev · 27/04/2026

An agent deleted his prod db with one API call Backups were tied to same volume Least privilege. Environment isolation. Secret management. Backup isolation. Audit logs. Recovery testing. None of this is new The agent didnt invent a new failure mode. It compressed an old one into nine seconds

330

Raphael De Lio @raphaeldelio.dev · 25/04/2026

Harness engineering is just ML folks hitting production for the first time, seeing their systems fail because they never had to think about security, scalability, reliability, maintainability, or system design as a whole, and then reinventing software engineering with a new name.

192

Raphael De Lio @raphaeldelio.dev · 18/04/2026

When auditors found out they couldn’t secure Access with all the data, they spent weeks cobbling together something that satisfied them. It’s like a free puppy… they say. Because it’s not the upfront capital that kills you, it’s the operations and maintenance on the backend.

010

Raphael De Lio @raphaeldelio.dev · 18/04/2026

They were referring to a database reporting tool that was implemented by a summer marketing intern. Given age only had a couple of months to work on it, it was actually pretty brilliant. It started being used in daily operations and consequently it was handed off to the IT department to support it

100

Raphael De Lio @raphaeldelio.dev · 18/04/2026

“It’s not the upfront capital that kills you, it’s the operations and maintenance on the backend.” Just read this in the book The Phoenix Project by @realgenekim.bsky.social , Kevin Behr, and George Spafford. Funny how actual it sounds.

220

Raphael De Lio @raphaeldelio.dev · 17/04/2026

Next Monday @iamsoham.bsky.social and I will be delivering two insightful talks at the Utrecht Java User Group. Join us to learn how AI agents can power autonomous DevSecOps pipelines and how memory works in both humans and AI Agents. www.meetup.com/nl-nl/utrech...

meetup.com

Meetup Utrecht JUG April 20th: Agentic Systems in Action, ma 20 apr 2026, 17:30 | Meetup

We are excited to announce our next Utrecht JUG meetup on **April 20, 2026,** hosted at the **Kamer van Koophandel.** Join us for an evening focused on the future of softwa

Raphael De Lio @raphaeldelio.dev · 17/04/2026

Always a pleasure to be back at @springio.net . Especially when talking about building agentic systems with Spring AI! #java @spring-ai.bsky.social

161

Is it suitable for agents that need to scale and serve hundreds or even thousands of users concurrently?

100

I’m seeing a lot of posts claiming n8n is dead. I’ve never used it and I wonder if it’s a tool that has value when building enterprise scalable agents? Or did it serve a small niche that was agglutinated by something else?

100

Well, mine just won't stop recommending me pizza. I never liked pIzza stuff on bsky... nonetheless I won't stop seeing posts from people sharing pics of the beautiful pizzas they make.

10/ Now, a an unrestricted key is a node is a much larger graph of trust relationships. 𝗜𝗳 𝗮 𝗺𝗮𝗹𝗶𝗰𝗶𝗼𝘂𝘀 𝘂𝘀𝗲𝗿 𝗰𝗮𝗻 𝘁𝗿𝗮𝘃𝗲𝗿𝘀𝗲 𝘆𝗼𝘂𝗿 𝗴𝗿𝗮𝗽𝗵, 𝗶𝘁 𝗯𝗲𝗰𝗼𝗺𝗲𝘀 𝗮 𝗽𝗮𝘁𝗵𝘄𝗮𝘆 𝘁𝗼 𝗺𝘂𝗰𝗵 𝗺𝗼𝗿𝗲 𝗰𝗮𝘁𝗮𝘀𝘁𝗿𝗼𝗽𝗵𝗶𝗰 𝗼𝘂𝘁𝗰𝗼𝗺𝗲𝘀.

100

9/ 𝗧𝗵𝗶𝘀 𝗶𝘀 𝗲𝘅𝗮𝗰𝘁𝗹𝘆 𝗵𝗼𝘄 𝗹𝗮𝘁𝗲𝗿𝗮𝗹 𝗺𝗼𝘃𝗲𝗺𝗲𝗻𝘁 𝗵𝗮𝗽𝗽𝗲𝗻𝘀 𝘄𝗶𝘁𝗵𝗼𝘂𝘁 𝘁𝗼𝘂𝗰𝗵𝗶𝗻𝗴 𝗮 𝘀𝗶𝗻𝗴𝗹𝗲 𝘀𝗲𝗿𝘃𝗲𝗿. A $70k bill is painful. But it’s contained and obvious. The multinational he works for can live with it.

100

8/ - Your Jira might contain customer secrets - Your artifact store might contain embedded credentials - Your support system might expose third-party access

100

7/ • Data plane keys (S3, logs, tickets) → expose internal data and often more credentials • Control plane / compute keys (CI/CD, artifact repos) → allow publishing, modification, and supply chain impact Once an attacker gets in, every integration becomes a pivot point.

100

6/ 4.Those credentials unlock other platforms (Jira, artifact stores, ticketing systems) 5.The process repeats across systems, accounts, and even companies The important detail is that not all keys are equal, but 𝗮𝗹𝗹 𝗸𝗲𝘆𝘀 𝗮𝗿𝗲 𝘂𝘀𝗲𝗳𝘂𝗹.

100