TLDR Dev 2026-04-15

turbopuffer is the serverless search engine for AI. It's built on object storage with smart caching - so it's fast, cheap, and extremely scalable.

✅ Used by Cursor, Notion, Linear, Cognition, Atlassian, Ramp, Granola, Legora...

✅ Fast → ~20ms p90 latency on 10M documents

✅ Truly serverless → Run your first query in ~4 minutes

✅ Up to 95% cheaper than traditional vector/text search databases

TLDR readers get their first month free. Sign up here.

Slack Engineering overcomes LLM context window limitations in long-running multi-agent security investigations by using three complementary context channels. This system uses a Director's Journal for structured memory, a Critic's Review to assess and prune inaccurate findings, and a Critic's Timeline to construct a plausible event chronology.

Relying only on scalar types creates subtle semantic bugs that bypass compilers because they cannot distinguish between different domain concepts, a risk not fully mitigated even by grouping parameters into structs. The solution is to wrap every meaningful domain concept in a distinct "newtype," ensuring the compiler enforces semantic correctness and leading to safer, self-documenting code with compile-time validation.

A developer created the custom static analysis linter "transactioncheck" using Go's `go/analysis` framework to resolve subtle database transaction bugs that caused data corruption and race conditions often missed by reviews and tests. This linter works by traversing the Abstract Syntax Tree to verify that all database operations correctly use the transaction-scoped parameter, efficiently catching existing violations and preventing future bugs at compile time by running in CI.

The hacker mindset involves seeing through a system's superficial rules and abstractions to understand its deeper, underlying mechanics, allowing for unconventional and highly efficient outcomes. This approach shows the power of understanding how a system truly operates.

The AI Coding Agent Manifesto establishes core values emphasizing contracts over conventions, verification over generation, and explicit communication to guide AI-driven software development. The manifesto then outlines twelve principles, such as humans defining code skeletons, documenting agent conventions, separating code generation from judgment, and treating AI-generated code as "guilty until proven innocent."

Clerk Billing now lets you cap organization membership based on which plan a customer subscribes to. Set a limit in the dashboard, and Clerk automatically blocks new invites once orgs hit it and prompts them to upgrade. No custom logic required.

Read the changelog

Plain is an opinionated Python web framework, forked from Django and redesigned for the "agentic era," that provides an explicit, typed, and predictable foundation for both human developers and AI agents.

Google Chrome is launching Skills, a new feature that allows users to save and reuse their most effective AI prompts as one-click tools for recurring tasks across different web pages. These personalized tools streamline workflows like product comparisons or document analysis.

Claude Code Routines automate unattended, repeatable tasks, such as code review or alert triage, by running saved configurations autonomously on Anthropic's cloud infrastructure. Users can trigger these routines via a schedule, an API call, or GitHub events like pull requests, and manage them through the web interface, CLI, or desktop app.

Open-source projects like Jellyfin face massive backlogs of unmerged PRs due to bottlenecks, such as a single maintainer nearing 100% utilization, which results in exponentially growing wait times and a "batch size death spiral" where large PRs clog the system. Combat this problem by optimizing the existing maintainer's time. Some solutions include capping PR size, gating quality, limiting work in progress, setting review cadences, and building a reviewer tier.

Government-funded academic research spanning 30 years, focused on parallel computing, graphics, and stream processing, directly led to the development of GPU computing, which is now vital for the modern artificial intelligence revolution and holds a lot of economic value. This foundational academic expertise was transferred to industry, evolving languages like Brook into CUDA at Nvidia.

An evaluation of Gemma 4 as a local agentic coding model in Codex CLI, compared against cloud models like GPT-5.4, showed that first-pass reliability, not raw token speed, was the most important performance metric, as cloud models were the fastest and most reliable. While cloud models still offer the best overall performance, Gemma 4's tool-calling ability makes local agentic coding viable.

Your benchmarks are in English. Are your users? Get native-language training data for 155+ language-locale pairs and 8 global regions. Keep your AI safe and reliable in the real-world contexts it actually ships in. See how it works

Multi-agentic software development is fundamentally a distributed systems problem with inherent impossibility results, like FLP and Byzantine Generals, that smarter LLMs alone cannot overcome, thus necessitating explicit coordination protocols and languages rather than waiting for future model improvements.

Traditional presentations of the Fifth Normal Form (5NF) are confusing due to unmotivated examples and illogical splitting, but designing schemas from business requirements and logical models naturally produces fully normalized physical schemas through patterns like the AB-BC-AC triangle or ABC+D star, making explicit 5NF decomposition unnecessary.

Figma's MCP server enables AI agents to accurately bridge the gap between design and code, and vice versa, by transforming raw design data into a structured, LLM-friendly format.

Google is making “back button hijacking,” where sites manipulate browser history to trap users, an explicit spam policy violation, with enforcement starting June 15.