When AI builds itself 👷, AI is not a line item 📝, local LLMs for agentic coding 🤖

AI development is transitioning to recursive self-improvement, where autonomous agents accelerate productivity in tasks like coding and research ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌  ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ 

Sign Up |Advertise|View Online TLDR Together With  TLDR Dev 2026-06-05 InsForge is the open-source backend infrastructure coding agents run themselves (Sponsor) Coding agents stop at code generation. The backend is the part they can't run themselves. InsForge gives them backend infrastructure to run, in the cloud or self-hosted. Here's what your coding agent can do with InsForge: > Database: create tables, run queries, and manage data. > Auth: add sign-in, sessions, and access control. > Storage: S3-compatible file storage. > Custom Compute: deploy and run services like a backend server, Redis, or Nginx. > Model gateway: call any LLM through one unified endpoint. Over 32K backend projects already run on InsForge, with 10K+ stars on GitHub. Check out the repo → 🧑‍💻 Articles & Tutorials Talking to JSI in Swift: what changed in SDK 56 (15 minute read) Expo SDK 56 has an architectural rewrite for native modules on Apple platforms by allowing Swift to communicate directly with JSI. This change removes the previous Objective-C++ middleware, resulting in performance gains between 1.6 and 2.3 times faster than older versions. By using Swift 5.9's C++ interoperability and non-copyable types, the new system has strict memory safety while simplifying the internal development stack. AI, Ashby Engineering, and the Future (17 minute read) At Ashby, over half of new production code is now AI-generated, yet software quality and stability remain consistently high. Engineers are required to have full accountability for their output by alternating between using AI as a collaborative sidekick for high-stakes decisions and a delegate for low-impact tasks. Sitar-agent: Building a reliable dynamic configuration sidecar at scale (13 minute read) Sitar-agent is a lightweight Kubernetes sidecar for reliably delivering dynamic configuration updates to thousands of service instances without redeployments, using Amazon S3 snapshots for fast bootstrapping and a periodic pull model for changes. The separate sidecar container helps with resource isolation and multi-language compatibility. 🧠 Opinions & Advice AI is not a line item (9 minute read) Treating AI expenditures as a single, monolithic budget line item based on token usage is a strategic mistake that discourages innovation. Instead, it's better to manage AI investments based on the specific functional value and experimental potential of individual tools. Single responsibility, the distorted principle (11 minute read) The Single Responsibility Principle (SRP) is often misunderstood as requiring modules to perform a single action. It actually recommends grouping cohesive behaviors under one responsibility. 🚀 Launches & Tools QVAC SDK: run local LLMs, now with 5x more context (Sponsor) Run local LLMs, speech, and image models fully on-device with QVAC, an open-source SDK. One npm install, no cloud, no API keys. The latest update integrated TurboQuant, unlocking up to 5x more context from the same machine, near-lossless. Get it on GitHub or see the docs at qvac.tether.io. Elixir v1.20 released: now a gradually typed language (9 minute read) Elixir v1.20 introduces a gradual, set-theoretic type system that performs automated type inference and checking, using a specialized dynamic() type to accurately identify bugs and dead code. The release also improves compilation speed and allows the compiler to detect potential runtime failures via type narrowing in guards and clauses. Boxes (Website) Boxes provides a cloud-based workbench that allows developers to run multiple AI coding agents in parallel within isolated, persistent virtual machines that can be monitored and managed from any device. 🎁 Miscellaneous When AI builds itself (29 minute read) AI development is transitioning to recursive self-improvement, where autonomous agents accelerate productivity in tasks like coding and research by automating their own refinement. While this offers benefits for discovery and healthcare, it also risks humans losing control. VoidZero is joining Cloudflare (12 minute read) VoidZero is joining Cloudflare to speed up the development of foundational web tools like Vite, Vitest, and Oxc. These projects will remain open-source and vendor-agnostic, so developers can continue building portable applications that run on any platform. Cloudflare is also committing a $1 million ecosystem fund to support maintainers and intends to transition its own developer tools to be built directly on top of Vite. Using local LLMs for agentic coding (26 minute read) Developers can use local LLMs with tools like LM Studio or Ollama for agentic coding to bypass rising cloud costs and improve privacy. Achieving optimal performance requires carefully configuring suitable models and hardware offloading settings, or using services like OpenRouter as an alternative for free hosted models. ⚡ Quick Links Why learn Pandas? (10 minute read) Pandas is a Python library that simplifies data analysis by treating datasets as DataFrames for efficient, loop-free calculations. Choosing Values for Robust Tests (5 minute read) Developing robust software tests requires using non-default, distinct, and varied input values to ensure code logic is correctly executed and to avoid false positives caused by coincidental default values. They're Made Out of Weights (5 minute read) Two observers grapple with the revelation that artificial intelligence functions through nothing but numerical weights and matrix multiplication, eventually deciding to dismiss its human-like behavior as mere pattern matching to avoid the ethical implications of its potential sentience. It is an amazing time for programmers! (3 minute read) Modern software developers live in a unique era where they can actively engage with and learn from legendary pioneers of computer science who remain accessible and active in the field. Love TLDR? Tell your friends and get rewards! Share your referral link below with friends to get free TLDR swag! https://refer.tldr.tech/d0af4569/3 Track your referrals here. Want to advertise in TLDR? 📰 If your company is interested in reaching an audience of web developers and engineering decision makers, you may want to advertise with us. Want to work at TLDR? 💼 Apply here, create your own role or send a friend's resume to jobs@tldr.tech and get $1k if we hire them! TLDR is one of Inc.'s Best Bootstrapped businesses of 2025. If you have any comments or feedback, just respond to this email! Thanks for reading, Priyam Mohanty, Jenny Xu & Ceora Ford Manage your subscriptions to our other newsletters on tech, startups, and programming. Or if TLDR Dev isn't for you, please unsubscribe.