Fullstack

Using LLMs to Automate Data Cleaning and Transformation Pipelines

A new paradigm, not a replacement of data engineering, but a fundamental shift in where engineering effort concentrates. If you were to ask a data engineer about their week, I am sure they would not speak about anything exciting. Most of their time is spent on data wrangling, messy upstream data, inconsistent date formats, null values that are not really null, and vendor exports that rename columns without noting those changes in the documentation.

03/06/2026

Multi-Scale Feature Learning in CNN and U-Net Architectures

Scale variation is a persistent source of error in vision models. A semantic concept can occupy a handful of pixels or most of the frame, and dense prediction tasks such as semantic segmentation intensify the difficulty because each output location must be both correctly classified and precisely localized.

03/06/2026

7 Technology Waves I’ve Seen in 30 Years of Software — Will AI Be the Next Real Transformation?

A Small Program and a Dot Matrix Printer In the early 1990s, one of the applications I worked on ran on a single PC in a small office. The program generated invoices and printed them on a dot matrix printer. The interface was text-based, the hardware was limited, and the system served only a handful of users. The application was built using Clipper and early PC-based database tools.

Microsoft bets the enterprise AI race will be won on data context, not model power

Microsoft used its Build 2026 developer conference on Tuesday to note that the hard part of enterprise AI is no The post Microsoft bets the enterprise AI race will be won on data context, not model power appeared first on The New Stack .

“A successful attack could be catastrophic”: Anthropic gives more groups access to Claude Mythos

Anthropic laid down some sobering words on Tuesday. “A successful attack on their codebase could be catastrophic. For most partners, The post “A successful attack could be catastrophic”: Anthropic gives more groups access to Claude Mythos appeared first on The New Stack .

Why Stable RAG Answers Can Still Hide Unstable Evidence

Most RAG evaluations focus on the answer. Is the answer correct? Does it appear grounded? Did retrieval metrics improve after a pipeline change? Those checks are useful. But they do not tell the full story.

How GitHub plans to win developers back

For much of the past year, GitHub has not been the stable utility developers had long gotten used to. Outages, The post How GitHub plans to win developers back appeared first on The New Stack .

Microsoft really, really, really wants developers to love Windows again

At its Build developer conference on Tuesday, Microsoft is launching so many new features for developers, it’s hard to keep The post Microsoft really, really, really wants developers to love Windows again appeared first on The New Stack .

With Intelligent Terminal, Microsoft is reinventing the Windows terminal

The good old terminal has been through a bit of a comeback in recent months, largely thanks to the popularity The post With Intelligent Terminal, Microsoft is reinventing the Windows terminal appeared first on The New Stack .

Microsoft debuts “Scout” at Build, a new personal agent for work

At Microsoft Build on Tuesday in San Francisco, the company debuted Microsoft Scout, a new personal agent for work that The post Microsoft debuts “Scout” at Build, a new personal agent for work appeared first on The New Stack .

When Snowflake Lies to You: Understanding False Failures in dbt Pipelines

The Problem Most Teams Get Wrong Every data engineer has lived this moment. A dbt model fails at 3 AM. You pull up the logs, see a type conversion error, and start digging through SQL. You check recent commits. Nothing changed. You inspect the upstream data. Nothing looks off. You rerun the job. It passes. You shrug, label it a transient issue, and go back to sleep.

Migrate a Hardcoded LangGraph Agent to LaunchDarkly AI Configs in 20 Minutes

In this tutorial, you’ll run a small LangGraph agent locally, then migrate its hardcoded prompts, model choice, and tools into LaunchDarkly AI Configs. After the migration, every prompt tweak, model swap, or tool change ships as a LaunchDarkly update instead of a code deploy. The migration takes about 20 minutes. When you finish, the codebase will:

Stop Debugging Glue Jobs Manually: Building an Agentic Observability Layer for Data Pipelines

The Pipeline Did Not Fail Cleanly Most pipeline failures don't look like "the job failed." Consider a common scenario. A Glue job reads overnight event files, applies business rules, and writes to an Iceberg curated table. The job runs at its scheduled time and errors out partway through. The control table shows SUCCESS for the previous batch and FAILED for the current one, which is what you'd expect.

Alternative Structured Concurrency

Java structured concurrency has been under development for a span of 5 years, weaving through 8 (!) distinct JEPs (JEP 428, JEP 437, JEP 453, JEP 462, JEP 480, JEP 499, JEP 505, JEP 525). To me, this feels rather excessive for what could be considered a fairly concise feature. My goal here is to experiment with an alternative approach that leverages Java's tried-and-tested, robust functionality available since JDK 1.5.

When One MVP Is Really Four Systems: A Better Way to Plan Multi-Role Apps

Teams often say they are building one app. A lot of the time, that is not true. I saw this while reviewing a telemedicine MVP. At first, the plan sounded simple enough: video visits, messaging, scheduling, and basic records.

5 AI Security Incidents That Broke Things in Production (and What They Have in Common)

Amazon's internal coding tool deleted a live AWS environment. A consulting firm's internal chatbot was fully compromised in two hours with no credentials. A calendar invite was enough to pull files off a developer's machine without a single user click.

OpenAI, Anthropic, Google, Amazon, and xAI all fail on type of attack, study finds

The safety benchmarks enterprise buyers rely on to evaluate AI models are measuring the wrong thing. That’s the finding from The post OpenAI, Anthropic, Google, Amazon, and xAI all fail on type of attack, study finds appeared first on The New Stack .

GenAI Implementation Isn't Magic — It’s a Lifecycle

If you’re planning to build an application using GenAI, it might seem like something completely new and complicated — but honestly, it’s not very different from building any other application. Just like any project, you still need a clear lifecycle and a proper framework to design, build, and successfully deploy your solution to production. Think of it like building a house. You don’t just start putting bricks together — you first plan, design, gather materials, and then build step by step.

Optimizing Databricks Spark Pipelines Using Declarative Patterns

If you've ever inherited a Spark job that runs in 35 minutes and someone asks you to make it faster, you know the routine. You start by checking partition counts, then file sizes, then shuffle stages, then broadcast hints. You find a handwritten OPTIMIZE schedule from 2022, a Z-ORDER on the wrong column, and a cluster sized for last year's data volume. By the time you've made the job fast, you've absorbed three new things to maintain. The next person to inherit it will absorb four.

Data Contracts as the "Circuit Breaker" for Model Reliability

Intro: When Good Models Go Wrong A few years ago, I spent months working on a microservices-based customer intake processing system for our application. The code was good, the tests were passing, and we had load-tested it with crazy high TPS.

MuleSoft IDP: Enhancing Efficiency and Accuracy in Data Extraction

This article will help developers, architects, and readers understand MuleSoft's Intelligent Document Processing capabilities and functionality. After reading this article, the reader will understand how to use MuleSoft Intelligent Document Processing and the different use cases where it can be used. MuleSoft Intelligent Document Processing (IDP) helps you read and understand documents like invoices, purchase orders, and other structured or unstructured files.

Jakarta EE 12: Entering the Data Age of Enterprise Java

For decades, Jakarta EE has addressed the challenge of building enterprise systems that endure technological change. The platform has evolved from monoliths to microservices, from application servers to Kubernetes, and from relational databases to distributed data platforms, all while maintaining its core strength: compatibility. Jakarta EE 12 marks another significant transition, shifting the focus beyond cloud-native infrastructure and APIs to prioritize data.

At Sapphire, SAP makes the case that enterprise AI is a context problem

At SAP Sapphire 2026 in Orlando in May, SAP made the case that the enterprise AI race will not be The post At Sapphire, SAP makes the case that enterprise AI is a context problem appeared first on The New Stack .

Gavriel Cohen found his own code inside OpenClaw, so he walked away

When Gavriel Cohen first saw OpenClaw, he knew he wanted it. At the time, Cohen (soon to be the founder The post Gavriel Cohen found his own code inside OpenClaw, so he walked away appeared first on The New Stack .

AI retrieval at scale is becoming a systems problem, not a tooling problem

AI retrieval has moved well beyond embeddings and vector search. Early retrieval architectures focused primarily on semantic similarity. Still, production The post AI retrieval at scale is becoming a systems problem, not a tooling problem appeared first on The New Stack .

The DIY platform trap that’s burning out engineering teams

Platform engineers are some of the most resourceful people in IT. Give them a problem, and they’ll automate their way The post The DIY platform trap that’s burning out engineering teams appeared first on The New Stack .

I tested Cursor’s new Jira integration and it’s 5 stars, no notes. Here’s why.

Cursor launched its Jira integration last week. The integration was marketed as simple: assign a ticket in Jira and Cursor The post I tested Cursor’s new Jira integration and it’s 5 stars, no notes. Here’s why. appeared first on The New Stack .

Why GPT-5.4, Claude, and Gemini can’t agree on basic, real-world facts

As the frontier model race accelerates, AI devotees are splitting their loyalty across the major providers at both the user The post Why GPT-5.4, Claude, and Gemini can’t agree on basic, real-world facts appeared first on The New Stack .

30/05/2026

Opus 4.8 Made Claude Smarter. Token Discipline Got Urgent.

I’m Matt Burns, Chief Content Officer at Insight Media Group. Each week, I round up the most important AI developments, The post Opus 4.8 Made Claude Smarter. Token Discipline Got Urgent. appeared first on The New Stack .

30/05/2026