Uday

I get the need for standardization, but through global governance like a W3C for AI

Merchants of complexity strike again with Google's A2A joining Anthropic's MCP. If AI is supposedly marching toward superintelligence, why are we creating more layers of abstraction instead of simplifying?

The protocol graveyard keeps growing. SOAP died. BPEL faded. ESBs collapsed under their weight.

LLMs should adapt to established standards, not the other way around.

Keep. It. Simple.

I'm not jumping on the MCP bandwagon just for LLM context. OpenAPI specs and REST APIs over Unix domain sockets already solve most of these problems effectively with mature tooling and adoption.

The promise of AI-assisted development should be simplification, not more protocols and abstractions.

Love this tiny thought from Shane Parrish

Everyone gets knocked down; the difference is who stays down.

When you’re on the ground, you’ll hear people saying, ‘That's not fair’ or ‘That shouldn't happen.’ The longer you lie there, the louder the voices become.

Getting up isn’t easy, but it’s the only way forward.

Sometimes the best engineering is hitting ⌫

Of course, there’s more to refactoring code than deletes.

As LLMs grow more capable at reasoning and specialized tasks, we should question our reflex to add more abstraction layers. The future of AI isn't about building more complex frameworks - it's about achieving more with less. True innovation often means elegant simplification
#ai #ml #kiss

💯 Adequate test coverage is a prerequisite for refactoring.

Started coding in college making simple tools. My first app just needed addition and if-else logic. Start small, focus on solving real problems - math will follow naturally when needed!

Great to see one of our customers already getting strong results with Brain Studio! Looking at their first week of data, they've maintained consistently high success rates (>95%) while processing over 500K tokens. Love seeing Brain Studio help developers ship with confidence! 💪

The rest? Getting stakeholders aligned, navigating compliance for sensitive data, and honestly just understanding what users actually needed vs what they said they wanted.

While everyone's obsessing over models that can "replace engineers", they're missing the real story. We just had a project where the tech was maybe 40% of the challenge.

Been seeing a lot of noise about O3 and the usual "end of SE" hot takes 🙄

But hey, here's something actually interesting that's flying under the radar - ModernBERT just dropped, and it's pretty sweet. 8x longer sequences and way faster processing!

huggingface.co/blog/modernbert

Try @deno.land or Bun too. Lot of innovation in this space

And please, spare me the 'but fraud protection!' sermon. Yes, it's nuanced, but 2-3% in 2024?

Just paid $195 for pest control... or wait, $201.96 because apparently my credit card is also a pest that needs controlling. 🙄 FedNow, where you at?
#InterchangeFees #DeathByAThousandFees

“Stop trying to be spectacular. Start being consistent.”
—Shane Parrish

💯 The best engineering teams I've worked with focused on being boringly reliable rather than trying to be super smart.

"Just a ChatGPT wrapper"
"Just a database wrapper"
"Just an AWS wrapper"
"Just a CRUD app"

Every meaningful software product wraps powerful primitives with purpose-built abstractions that solve real problems. The value is in how you orchestrate and deliver.

#buildinpublic #buildwithai

During yesterday's OpenAI outage, Brain Studio's observability caught the disruption immediately. Our fault tolerance kicked in - automatic retries and fallbacks kept customer systems running. Solid engineering practices matter - especially with AI.
#BuildWithAI #BuildInPublic

Languages as personalities:

Java: Corporate vet who memorized design patterns

Go: Efficiency expert optimizing morning routines

Rust: Safety inspector requiring permits for everything

TypeScript: Java's cousin who does frontend

Python: Prof who knows everyone

(Java dev for 15+ years 🙂)

How do you know if your LLM prompts are actually working?

We built internal tools to find out. Now making them open source. Brain Studio helps you track prompt performance, catch regressions, and optimize costs - across any LLM provider. First preview next week.

#BuildWithAI #BuildInPublic

AGI probably already exists but it's trapped in an infinite loop trying to find the 'Submit' button on a government site 🙂

Repository and documentation will be available soon. If you're interested in early design discussions or contributing, especially around telemetry integration, let's connect.

4/4
We're currently using this library with our DECSEA (Deterministic Code Synthesis via Execution Analysis) system, which successfully generates 30-50% of its own implementation code through controlled LLM interactions.

#LLMs #Java #OpenSource #buildinpublic

3/4
The library aims to solve common integration challenges:
• Minimalist and easy to use
• Support virtual threads
• Structured telemetry for LLM operations
• Seamless error handling and retry mechanisms

2/4
Technical specifications:
• Built on JDK's HTTP client with virtual thread support
• Core dependencies limited to JSON parsing and OpenTelemetry
• GraalVM integration for Python interoperability
• Native image compatible

1/4
After evaluating Java libraries for LLM integration, we identified a need for a minimalist solution focused on modern Java features. While comprehensive libraries like Langchain4j serve many use cases well, we're developing an alternative that prioritizes simplicity and performance.

💯 AI can potentially simplify ERP customizations

4/4 No more stitching together dozens of SaaS tools or building complex workflows. This isn't incremental—it's a fundamental rethinking of how we build & deploy business software.
#buildinpublic