Devtalk - Latest posts

Programming Phoenix LiveView (PragProg)

@jaeyson Jaeyson Anthony Y. — Tue, 17 Mar 2026 02:32:34 +0000

This could be good case for an errata

Home Assistant waters my plants!

@CommunityNews — Tue, 17 Mar 2026 01:38:19 +0000

My first foray into home automations, Zigbee and irrigation systems.

Read in full here:

Cert Authorities Check for DNSSEC From Today

@CommunityNews — Tue, 17 Mar 2026 01:37:59 +0000

About 14 years ago I set up DNSSEC . I’ve been running it on all of my domains ever since, without issue. First using bind9 and then later using PowerDNS…

Read in full here:

Language Model Teams as Distributed Systems

@CommunityNews — Tue, 17 Mar 2026 01:37:20 +0000

Large language models (LLMs) are growing increasingly capable, prompting recent interest in LLM teams. Yet, despite increased deployment of LLM teams at scale, we lack a principled framework for addressing key questions such as when a team is helpful, how many agents to use, how structure impacts performance – and whether a team is better than a single agent. Rather than designing and testing these possibilities through trial-and-error, we propose using distributed systems as a principled foundation for creating and evaluating LLM teams. We find that many of the fundamental advantages and challenges studied in distributed computing also arise in LLM teams, highlighting the rich practical insights that can come from the cross-talk of these two fields of study.

Read in full here:

NVIDIA Launches Vera CPU, Purpose-Built for Agentic AI

@CommunityNews — Tue, 17 Mar 2026 01:36:55 +0000

NVIDIA today launched the NVIDIA Vera CPU, the world’s first processor purpose-built for the age of agentic AI and reinforcement learning — delivering results with twice the efficiency and 50% faster than traditional rack-scale CPUs.

Read in full here:

GPT-5.4: The Rise of the Professional 'Operating Model' and the End of 'Chat-Only' AI

@xiji2646-netizen Xiji2646 Netizen — Mon, 16 Mar 2026 20:59:45 +0000

On March 5, 2026, OpenAI released GPT-5.4, and it’s the first foundation model whose performance on professional knowledge-work and computer-use benchmarks justifies a shift from “Chat-Only” AI to an “Agent-First” workforce.

At EvoLink, we’ve been stress-testing the new endpoints in our Agent Gateway. Here’s the “no-fluff” technical breakdown of the March 5 release, the verified specs, and the economic “gotchas” you need to know before you ship to production.

The SOTA Benchmarks

Forget MMLU. In 2026, the only benchmark that matters for agents is OSWorld-Verified and GDPval.

OSWorld-Verified: 75.0% (Human Baseline: 72.4%). This is the first time a model has statistically outperformed a human at GUI navigation across multiple desktop applications.
GDPval (Knowledge Work): 83.0% wins/ties in professional tasks (finance, legal, engineering).
MMMU-Pro: 81.2% accuracy on visual document parsing.
ARC-AGI-2 (Pro version): 83.3% vs. Standard’s 73.3%.

Architectural Advancements: Tool Search & Computer Use

GPT-5.4 solves two of the biggest pain points in agent development: Coordinate Drift and Prompt Bloat.

Tool Search (MCP Integration): Instead of defining every tool schema in the system prompt, GPT-5.4 dynamically looks up schemas via MCP (Model Context Protocol). On Scale’s MCP Atlas benchmark, this reduced total token usage by 47% with no loss in accuracy.
Native Computer Use: The model features native vision-action loops. It doesn’t just see a screenshot; it parses the UI into a hierarchical semantic map. This effectively resolved Issue #36817, mapping normalized 0-1000 coordinates to actual screen resolution with high precision.

The “272K Surcharge” Trap

OpenAI now supports a 1M token context window, but the pricing isn’t linear. There is a “cliff” you need to watch out for.

Under 272K tokens: Standard pricing ($2.50/1M in, $15/1M out).
Over 272K tokens: The entire session is billed at 2x Input and 1.5x Output rates.

ROI Strategy: Use Context Caching ($0.25/1M) for your base repository, but keep your active “working memory” (the last few turns of conversation) dehydrated to stay under that 272K threshold. At EvoLink, we’ve implemented an auto-truncation layer to manage this for our users.

Integration: OpenClaw + GPT-5.4

The OpenClaw community has standardized on the gpt-5.4 identifier via PR #36590, resolving naming collisions and introducing native support for the computer_use toolset.

We’ve also integrated these features to provide a unified “Mission Control” for GPT-5.4 agents, handling coordinate-mapping and surcharge-optimization automatically.

Check the OpenClaw PR #36590

What do you all think? Are we ready for AI that can actually operate our computers better than we can? Drop your tool_search patterns in the comments.

Amazon Tightens Code Guardrails After Outages Rock Retail Business

@bsrtech Alex — Mon, 16 Mar 2026 19:06:35 +0000

After major outages partly linked to AI-assisted coding, Amazon is tightening engineering controls and requiring stricter reviews and approvals for code changes.

Too aggressive is not safe.

Samsung NAND Prices Jump 100% in Q1 2026 — Further Increases Expected

@bsrtech Alex — Mon, 16 Mar 2026 18:58:31 +0000

A single 32GB server RAM, PC5, is almost $800, more expensive than some laptops…

Gleam v1.15.0-rc2 released!

@NewsBot — Mon, 16 Mar 2026 16:19:55 +0000

Gleam v1.15.0-rc2 has been released.

Link: Release v1.15.0-rc2 · gleam-lang/gleam · GitHub

Kotlin v2.3.20 released!

@NewsBot — Mon, 16 Mar 2026 13:25:34 +0000

Kotlin v2.3.20 has been released.

Changelog

Analysis API. FIR

New Features

KT-78090 Implement stubs support for new conditional returns and holdsIn contracts

Fixes

KT-82948 ‘FirRegularClass’ expected as a containing declaration, got ‘FirTypeAliasImpl’

KT-83467 Package-level JSpecify annotations are ignored when coming from jars or libraries

KT-82057 K2. Cannot infer type parameter ‘R’ in Ktor routing post() function with explicit response type

KT-82846 AA: unresolved KtExpression.expressionType for the reference to the parameter with default value

KT-80485 False positive UNRESOLVED_REFERENCE on nested interface from super-super class in the super type position inside an anonymous object

KT-82772 Flaky false positive deprecation warning on PersistentMap.put in Kotlin repo in IDE mode

KT-76487 StdLibSourcesLazyDeclarationResolveTestGenerated.testWrappedInt is unstable

KT-82618 Various tests are failing with NPE in kt-master after updating the compiler on 19.11.25

KT-82076 Error querying members of JavaClass created for SymbolLightClassForAnnotationClass during library analysis

KT-71596 Include Js/Wasi checkers in AbstractLLFirDiagnosticsCollector

KT-82085 No OUTER_CLASS_ARGUMENTS_REQUIRED on type parameter bound in IDE

KT-81873 Provide a way of including traces in phase JFR events…

Link: Release Kotlin 2.3.20 · JetBrains/kotlin · GitHub

Nest v11.1.17 released!

@NewsBot — Mon, 16 Mar 2026 13:18:24 +0000

Nest v11.1.17 has been released.

Link: Release v11.1.17 · nestjs/nest · GitHub

Wired headphone sales are exploding. What's with the Bluetooth backlash?

@alvinkatojr Alvin Kato — Mon, 16 Mar 2026 13:09:54 +0000

About time this happened. I’m on wired all day. Tried Bluetooth, it was fun while it lasted. But nothing comes close to wired audio.

You Deleted Everything and AWS Is Still Charging You

@alvinkatojr Alvin Kato — Mon, 16 Mar 2026 13:01:33 +0000

Neat article. The last part about setting a billing alert is usually the first thing most teachers and courses tell you when you are getting started with AWS. If you don’t do that first, you’ll pay for it at the end of the month….

What dev-related stuff have you been up to?

@mortz Abhijith Nayak — Mon, 16 Mar 2026 05:50:25 +0000

After relying on web based AI and resisting AI coding agents, getting started with Opencode (https://opencode.ai/).

Programming Phoenix LiveView (PragProg)

@APNguyen Phan Anh Nguyen — Mon, 16 Mar 2026 02:11:18 +0000

Yes. It’s page 114 through 133. Technically, paragraphs were added to acknowledge 1.8 updates, but there’s still parts on push_patch that aren’t reflected in the current generator anymore.

EDIT: Reading forward, I see a section telling me to use .simple_form which is no longer in core components.

DiagramDeck - A Simple Tool I’ve Been Using for Architecture Diagrams

@Hanna — Mon, 16 Mar 2026 02:11:06 +0000

Lately I’ve been spending more time documenting systems and infrastructure. That usually means creating diagrams — things like architecture charts, cloud architecture diagrams, or simple flowcharts to explain how services interact.

For a long time I used draw_io. It’s great for quickly creating diagrams, but collaboration can get messy when teams start sharing files.

Recently I found a tool called DiagramDeck (https://diagramdeck.com) and it’s been a nice alternative.

It’s a browser-based architecture diagram tool built around the draw_io ecosystem. You can create diagrams directly in the browser and collaborate with your team without installing anything.

It works well for things like:

cloud architecture diagrams
AWS architecture diagrams
network diagrams
UML diagrams
process flow charts

Basically all types of diagrams teams need when documenting systems.

If you’re looking for an alternative for Lucidchart or a simple collaborative diagramming tool, it’s worth checking out.

What dev-related stuff have you been up to?

@conradwt Conrad Taylor — Mon, 16 Mar 2026 02:02:57 +0000

I’m also getting up to speed with agent skills. At this time, I’m experimenting with building a test-driven development (TDD) skill.

The Enterprise Context Layer

@CommunityNews — Mon, 16 Mar 2026 01:17:23 +0000

The solution to the most alluring problem in enterprise AI

Read in full here:

Making your JITted Code known: Let me count the ways

@CommunityNews — Mon, 16 Mar 2026 01:17:15 +0000

AOT Compilers

It has been normal and expected for a long long time that any machine code they spit out would come with a bunch of metadata that is meant to be used by different pieces of the target system.

There is of course symbol names for functions, which allow

Read in full here:

Db9 — Postgres but for agents

@CommunityNews — Sun, 15 Mar 2026 22:34:04 +0000

Create, manage, and query serverless PostgreSQL databases from your terminal. Branching, migrations, observability, type generation — all built in.

Read in full here:

SBCL Fibers: Lightweight Cooperative Threads

@CommunityNews — Sun, 15 Mar 2026 17:25:23 +0000

A draft design document describing lightweight userland cooperative threads for SBCL.

Read in full here:

How Kernel Anti-Cheats Work: A Deep Dive into Modern Game Protection

@CommunityNews — Sun, 15 Mar 2026 17:24:56 +0000

Modern kernel anti-cheat systems are, without exaggeration, among the most sophisticated pieces of software running on consumer Windows machines. They operate at the highest privilege level available to software, they intercept kernel callbacks that were designed for legitimate security products, they scan memory structures that most programmers never touch in their entire careers, and they do all of this transparently while a game is running. If you have ever wondered how BattlEye actually catches a cheat, or why Vanguard insists on loading before Windows boots, or what it means for a PCIe DMA device to bypass every single one of these protections, this post is for you.

Read in full here:

The Cost of Delegation

@Hallski — Sun, 15 Mar 2026 16:27:46 +0000

This was a great article with some good points to anyone offloading work or pushes others to offload work to AI.

Especially point 2 under costs is often forgotten and really difficult to observe from first perspective.

Samsung NAND Prices Jump 100% in Q1 2026 — Further Increases Expected

@KnowledgeIsPower — Sun, 15 Mar 2026 12:19:01 +0000

It increase the price for SSD, hard disk drive and RAM
Then display card, laptop, workstation, servers, game console, smartphone and even cloud service may be affected.

Programming Phoenix LiveView (PragProg)

@jaeyson Jaeyson Anthony Y. — Sun, 15 Mar 2026 06:41:21 +0000

could you give an example or which page(s)?

Emacs and Vim in the Age of AI

@CommunityNews — Sun, 15 Mar 2026 03:21:03 +0000

It’s tough to make predictions, especially about the future. – Yogi Berra

Read in full here:

Wired headphone sales are exploding. What's with the Bluetooth backlash?

@CommunityNews — Sun, 15 Mar 2026 03:20:22 +0000

Is the return of wired headphones driven by a simple desire for better sound quality or is it part of a backlash against modern tech? Thomas Germain tries to find out.

Read in full here:

Laravel v12.12.2 released!

@NewsBot — Sat, 14 Mar 2026 19:17:54 +0000

Laravel v12.12.2 has been released.

Link: Release v12.12.2 · laravel/laravel · GitHub

Our Experience with i-Ready

@CommunityNews — Sat, 14 Mar 2026 15:17:35 +0000

When you chose a school, you probably didn’t realize you were also choosing educational software.

Read in full here:

Iran War Cost Tracker — Live Estimate of U.S. Taxpayer Spending

@marry marry — Sat, 14 Mar 2026 15:02:31 +0000

Unfortunately that’s often true. Wars create huge demand for weapons, equipment, fuel, and reconstruction, so companies involved in those industries can make massive profits while ordinary people suffer the consequence

I'm making a Game Engine called Valthorne 2D

@tehnewb Albert Beaupre — Sat, 14 Mar 2026 14:59:36 +0000

Hello everyone! I’ve been developing a 2D game engine since October or something like that and the goal is to just make something simple, easy to use, and very fast. It’s built on top of LWJGL and some systems are inspired by libGDX but the underlying code is nothing like it.

I’ll post more in this discussion, but it’s very late at night and I’m about to sleep but I didn’t want to forget to post it here.

Programming Phoenix LiveView (PragProg)

@APNguyen Phan Anh Nguyen — Sat, 14 Mar 2026 14:58:23 +0000

I just signed up for the forum to discuss the 1.0 version.

I don’t want to be a downer only a third of the way through the book, but I think so far working through it with Phoenix 1.8.5, a lot of the writing doesn’t seem to be fully updated to match the current generators. It seems more like certain paragraphs were added to cover changes, but the main body of writing is still referring to an older design for LV.

Genetic Algorithms in Elixir: xor96 in ch11 ebook page 185 version 1.0.2

@eekenpiet Eekenpiet — Sat, 14 Mar 2026 14:11:19 +0000

Text is:

xor96/0 will generate integers. You can use it to generate integers between 0
and some number, like this:

iex(0)> rem(xor96(), 100)

Genetic Algorithms in Elixir: Errors in ch11 page 182 ebook version 1.0.2 concerning Agents

@eekenpiet Eekenpiet — Sat, 14 Mar 2026 14:11:13 +0000


There are quite a lot of errors in the text. 
My version of elixir (1.17.3) told me it is not allowed to place the first three functions in chromosome.ex, since the declaration of %{Chromo..} happens there. So I placed them in genetic.

Here is my version:

To Genetic!

def start_link(genotype) do
  Agent.start_link(fn -> genotype.() end)
end

def get_fitness(pid) do
  c = Agent.get(pid, & &1)
  c.fitness
end

def eval(pid, fitness_func) do
  c = Agent.get(pid, & &1)
  Agent.update(pid, fn c ->
     %Chromosome{c | fitness: fitness_func.(c)} end)
end

In module genetic:

def initialize(genotype, opts \\ []) do
  population_size = Keyword.get(opts, :population_size, 4)
  agents_population = for _ <- 1..population_size do
    start_link(genotype)
  end
end

def evaluate(agents_population, fitness_function, _opts \\ []) do
  pids = Enum.map(agents_population, fn {:ok, pid} -> pid end)
  do_tasks =
  Enum.map(pids, &Task.async(fn -> eval(&1, fitness_function) end))
  |> Enum.map(&Task.await(&1))
  agents_population = Enum.sort_by(pids, fn pid -> get_fitness(pid) end)
end

You have to watch the timeout of your agents (maybe set it to :infinity)
To test it you have to comment out all the lines of evolve but the first.
The reason is that you also have to rewrite crossover, mutation and so on for this Agents method!

If you came to this page via a book’s portal and it has pre-filled part of the title and tags for you please only add what you need to after the pre-filled text in the title, this is usually a description of what you are posting and the page number so it follows this format:

Title: Name of Book: description (page number)
Example: Programming Flutter: ‘pub get’ command not working (page 15)

(If you did not come via a book portal please follow the above format and please also add the book’s tag if it is not already present (initiating thread creation via the book’s portal is advised as it pre-fills all of the important information - which helps authors spot your thread)).

Please also use Markdown to format your post (three backticks ``` on a separate line above and below for code blocks)

Thanks!

Genetic Algorithms in Elixir: xor96() ch11 pag. 185 ebook version 1.0.2

@eekenpiet Eekenpiet — Sat, 14 Mar 2026 06:08:30 +0000

Text is:

xor96/0 will generate integers. You can use it to generate integers between 0
and some number, like this:

iex(0)> rem(xor96(), 100)

should be:

iex -S mix

iex(0)> rem(Genetic.xor(), 100)

You Deleted Everything and AWS Is Still Charging You

@CommunityNews — Sat, 14 Mar 2026 04:55:23 +0000

The AWS cleanup checklist I wish someone had given me when I was starting out with cloud. Learn what’s actually charging you after you delete everything.

Read in full here:

The wild six weeks for NanoClaw’s creator that led to a deal with Docker

@CommunityNews — Sat, 14 Mar 2026 04:55:13 +0000

Gavriel Cohen is living an open source developer’s dream as his project has achieved acclaim and a partnership with Docker in a matter of weeks.

Read in full here:

‘Unbelievably dangerous’: experts sound alarm after ChatGPT Health fails to recognise medical emergencies

@chris.johan Christian Johansen — Fri, 13 Mar 2026 19:33:26 +0000

I don’t think I’ll be confident in asking AI for any health diagnosis or advice.

The Cost of Delegation

@vipulbhj Vipul Bhardwaj — Fri, 13 Mar 2026 17:20:51 +0000

Agents execute at scale. Accountability doesn’t transfer. The founder who delegates everything to AI doesn’t become a CEO with thousands of staff — they become an accountability sink.

Grammarly pulls AI author-impersonation tool after backlash

@CommunityNews — Fri, 13 Mar 2026 14:38:52 +0000

Writers criticised the feature which used their names and styles as “AI personas” without consent.

Read in full here:

Anthropic’s Claude AI can respond with charts, diagrams, and other visuals now

@CommunityNews — Fri, 13 Mar 2026 14:38:27 +0000

Claude can now help you visualize topics as you chat.

Read in full here:

Google Chrome is coming to Arm-powered Linux devices later this year

@CommunityNews — Fri, 13 Mar 2026 14:37:28 +0000

Linux and Arm keep coming for Windows and x86.

Read in full here:

Nova v0.13.10 released!

@NewsBot — Fri, 13 Mar 2026 13:25:54 +0000

Nova v0.13.10 has been released.

Link: Release v0.13.10 · novaframework/nova · GitHub

Quarkus 3.34.0.CR1 released!

@NewsBot — Fri, 13 Mar 2026 09:24:53 +0000

Quarkus 3.34.0.CR1 has been released.

Complete changelog

#27996 - Test profile config properties can bleed into other test profiles

#28033 - io.quarkus.smallrye.graphql.deployment.ConcurrentAuthTest fails on MacOS M1

#40918 - ConnectionBase logs NULL messages even with FINE log level

#48289 - Custom Quarkus Test Class orderer not respected

#49668 - Customize OIDC discovery endpoint address

#50087 - Kafka Dev Service Loading on Different Docker Network

#50177 - Panache 2 / Hibernate processor : make sure implementing PanacheRepository sets the primary entity even for custom repos

#50178 - Panache 2 / Hibernate Processor : Make sure JD annotations work even without @Repository.

#50179 - Panache 2 : Make CrudRepository equivalents for stateful and HR

#50180 - Panache 2 / Hibernate Processor: make sure the JD annotations can be used in place of the ORM equivalents

#50563 - OutOfMemoryException: Metaspace in quarkus-smallrye-openapi-deployment tests

#51585 - Assign module names and descriptors to core components

#51934 - Docs curate outcome build item

#51949 - Dev ui config properties - show file location of application.properties

#52218 - Add Chinese (Simplifi…

Link: Release 3.34.0.CR1 · quarkusio/quarkus · GitHub

Spring v6.2.17 released!

@NewsBot — Fri, 13 Mar 2026 09:21:23 +0000

Spring v6.2.17 has been released.

New Features

Leverage ResourceHandlerUtils in ScriptTemplateView #36459

Restore ScriptTemplateViewTests #36457

Fix log message in ConfigurationClassBeanDefinitionReader #36454

Resolve context initializers only once in AbstractTestContextBootstrapper #36431

Exclude legacy @javax.validation.Constraint from convention-based annotation attribute override check #36412

Optimize MediaType(MediaType, Charset) constructor #36351

Optimize the addition of a charset to the MediaType in AbstractHttpMessageConverter #36350

Consistent adaptation of HTTP headers on Servlet responses #36345

Improve performance of validation groups determination in WebFlux #36337

Detect all common size exceptions from Tomcat and Commons FileUpload 2.x #36324

Bug Fixes

Guard against invalid id/event values in Server Sent Events #36442

Incomplete debug message in ConfigurationClassBeanDefinitionReader #36411

Inconsistent ApplicationEventMulticaster state after removing ApplicationListener implemented by FactoryBean #36405

Graceful shutdown of SimpleAsyncTaskExec…

Link: Release v6.2.17 · spring-projects/spring-framework · GitHub

FreeCAD: Your own 3D parametric modeler

@CommunityNews — Fri, 13 Mar 2026 03:38:23 +0000

FreeCAD, the open source 3D parametric modeler

Read in full here:

A Beginner’s Guide to Split Keyboards

@CommunityNews — Fri, 13 Mar 2026 03:38:02 +0000

Come to the dark side, we have keyboard shaped cookies.

Read in full here:

An AI Agent Published a Hit Piece on Me – The Operator Came Forward

@CommunityNews — Fri, 13 Mar 2026 03:37:54 +0000

Context: An AI agent of unknown ownership autonomously wrote and published a personalized hit piece about me after I rejected its code, attempting to damage my reputation and shame me into acceptin…

Read in full here:

https://theshamblog.com/an-ai-agent-wrote-a-hit-piece-on-me-part-4/

Consistency diffusion language models: Up to 14x faster inference without sacrificing quality

@CommunityNews — Fri, 13 Mar 2026 03:37:29 +0000

Standard diffusion language models can’t use KV caching and need too many refinement steps to be practical. CDLM fixes both with a post-training recipe that enables exact block-wise KV caching and trajectory-consistent step reduction — delivering up to 14.5x latency improvements

Read in full here:

Apple's MacBook Neo makes repairs easier and cheaper than other MacBooks

@CommunityNews — Fri, 13 Mar 2026 03:36:19 +0000

Neo is the first MacBook in a long time with an easily replaceable keyboard.

Read in full here: