Hacker News — 2026-05-17#

Top Story#

When Fisker went bankrupt, they left 11,000 Ocean SUV owners with $70k vehicles that were rapidly becoming rolling paperweights as the company’s cloud servers went dark. Instead of accepting the loss, an organized collective of 4,000 owners reverse-engineered the proprietary software patches, mapped the CAN buses, built Home Assistant integrations, and essentially stood up an open-source car company from the ashes. It’s a massive, tangible win for the Right to Repair movement and a damning indictment of the “software-defined vehicle” architecture that ties critical functionality to a startup’s fragile runway.

Front Page Highlights#

C++26 Shipped a SIMD Library Nobody Asked For A brutal, deeply technical teardown of std::simd graduating into the C++26 standard. The committee spent a decade polishing a library-based abstraction that arrived completely obsolete. Because the optimizer can’t see through the template abstraction layer, it compiles 10x slower and actually runs 2.4x slower than a plain auto-vectorized scalar for-loop. Worse, it completely ignores the cross-lane operations (shuffles, permutes) that make up 90% of real-world SIMD code. If you need portable SIMD, Google Highway or ISPC remain vastly superior choices.

Every AI Subscription Is a Ticking Time Bomb for Enterprise AI labs are currently running a massive loss-leader scheme, selling $20/month subscriptions that often cost $200-$400/month to serve via API equivalents. This was merely bad economics until agentic loops (like Claude Code) started burning through tokens autonomously, collapsing the flat-fee model. As companies like Anthropic and OpenAI push toward IPOs, public market pressure will force a harsh price correction—leaving enterprises that treated cheap AI as a utility scrambling to cover six-figure budget overruns.

Americans Are Smashing Flock Cameras A wave of vigilante hardware destruction is targeting Flock Safety’s ALPR (Automated License Plate Reader) cameras across the US. The backlash is driven by data showing local police using the network to funnel thousands of lookups to ICE, directly contradicting Flock’s claims. With city councils frequently ignoring overwhelming public opposition to contract renewals, residents are bypassing official channels and taking down the infrastructure with vice grips and metal cutters.

Native all the way, until you need text A 20-year veteran of macOS and iOS development waves the white flag on native text rendering. After battling SwiftUI’s limitations with document selection, CPU spikes with NSTextView streaming, and the immense difficulty of implementing basic features like dictionary lookup and context menus in pure TextKit 2, they concede that WebKit and Electron are vastly superior for rich text and markdown. It’s a sobering reminder of why modern chat-heavy interfaces inevitably gravitate toward the DOM.

Playing Atari ST Music on the Amiga with Zero CPU Peak demoscene wizardry. To beat a rival’s sin-dots rendering record, a developer needed to emulate the Atari YM2149 sound chip on an Amiga without spending any 68000 CPU cycles. By cleverly abusing the Amiga PAULA chip’s obscure “attached voice” modulation feature and pre-generating thousands of self-chaining COPPER lists, they managed to offload the entire audio emulation to the coprocessor. The result is a perfect “MadMax Buzzer” synth sound that runs at literally 0% CPU.

Hosting a website on an 8-bit microcontroller A classic “because I can” hardware hack. An AVR64DD32 microcontroller is too slow to generate a 10BASE-T Ethernet signal, so the author used the ancient Serial Line Internet Protocol (SLIP) over a 5V serial adapter. To make it publicly accessible without a dedicated IPv4 address, they routed it through a Wireguard tunnel on a Linux VPS.

Show HN & Launches#

• Semble: A code search library specifically built for AI agents. It uses a hybrid of BM25 and static Model2Vec embeddings to return exact code snippets in ~1.5ms on CPU, using 98% fewer tokens than the standard grep+read approach.
• Codiff: A fast, native macOS desktop diff viewer for reviewing staged and unstaged Git changes, featuring an LLM-generated walkthrough mode.
• rk3562deb: A from-scratch reverse engineering project that turns an $80 Doogee U10 Android tablet into a fully functional Debian 12 Linux workstation.

Discussion & Debate#

A meta-theme dominated the comments today around the true cost and utility of LLMs in the engineering loop. Sparked by Addy Osmani’s warning that developers are accumulating “cognitive debt” by letting AI agents solve problems without understanding them, the community is fiercely debating where AI actually saves time. This dovetailed perfectly with a pragmatist pointing out that AI doesn’t speed up processes when the real bottleneck is upstream requirements gathering. Meanwhile, local-model advocates are crunching the numbers, noting that running models on Apple Silicon amortizes to ~$1.50 per million tokens—making local hardware roughly 3x more expensive than subsidized cloud APIs like OpenRouter.