Study Finds AI-generated Code Fails to Boost Productivity Due to Downstream Bottlenecks
Harvard study of 718 firms (Jellyfish data): coding agents raise code volume, but completed work doesn't keep pace because of downstream bottlenecks.
AI Daily Recap
45 articles · 5 categories
The finishable daily brief
What happened in AI — Oct 22, 2026
Thursday, Oct 22, 2026
45 articles · 5 categories
read top to bottom · then stop
In 30 seconds
- Harvard study of 718 firms: AI coding raises code volume, not completed work.
- OpenAI's Patch the Planet helped find PHP2Shell flaws in core PHP.
- DeepSeek reportedly raised about $15B, led by CATL and Tencent, after a US federal ban.
- Modal open-sources Spindle for fine-tuning open models.
- Anthropic donates Agent Skills to the Agentic AI Foundation.
AI coding agents are producing more code without more finished work, according to a 718-firm study, even as new agent runtimes and gates keep arriving.
DeepSeek reportedly closed a $15B round days after the US ordered federal agencies to remove its products.
Coding agents: more code, same throughput 4 items
A 718-firm study finds AI-assisted code volume rises without a matching rise in completed work, while builders ship new embeddable and gated agent runtimes.
Vercel Labs Open-Sources fx: A Zig Coding Agent Designed to Be Embedded
Vercel Labs' experimental fx is a Zig coding agent that runs as a CLI, script component or embedded library, including a WebAssembly build.
The nightly Ralph loop: a coding agent that works my backlog and never merges
A nightly coding-agent loop works a backlog and never merges, leaving the merge decision to a human.
Developer Behind the AI-Built Adobe Clones Says They Are “Not Anywhere Close to Ready
A developer used Claude Opus 5.5 to build open-source stand-ins for seven Adobe apps, then said they are not close to ready after projecting near parity in a month.
Security: agents as both attack surface and bug finders 2 items
OpenAI's program helped find PHP core vulnerabilities, while agent-side defenses focus on local policy checks.
[AINews] OpenAI’s Patch the Planet program is 1 of 2 parties to discover PHP2Shell vulnerabilities in core PHP affecting 70% of the web, using GPT-5.6 Sol; coordinated response over 2 months
OpenAI's Patch the Planet program was one of two parties to discover PHP2Shell vulnerabilities in core PHP, using GPT-5.6 Sol, with a two-month coordinated response.
Show HN: Secure your AI agents on Mac
Plango is a Mac-native security tool that runs a local model to evaluate coding-agent actions against the lethal trifecta of data access, outside communication and untrusted content.
Inference and training infrastructure for agents 4 items
Teams are trimming latency and cost in agent serving, and open training APIs are reaching smaller teams.
Why AI Compute Comes in Rectangles — Erik Bernhardsson, Modal CEO
Modal's CEO discusses elastic GPUs, sandbox growth and the economics of the agent cloud.
Introducing our Training API
Modal's Spindle is an open-source API for fine-tuning open models; one command deploys a server that runs Tinker-compatible training code.
Stream low-latency speech from AI agents using Kokoro TTS on Amazon SageMaker AI
AWS shows streaming each sentence of an agent's reply to a Kokoro-82M TTS model on SageMaker AI so speech starts before the full response is ready.
Expanding Claude availability and deployment choice on Google Cloud
Google Cloud expands Claude availability and deployment options for developers.
Agent standards and platform APIs 2 items
Vendors are pushing shared agent-skill packaging and new low-latency API surfaces.
Donating Agent Skills to the Agentic AI Foundation
Anthropic donates Agent Skills to the Agentic AI Foundation.
Build more responsive apps with the Decisions API
OpenAI's Decisions API targets fast, low-cost decisions inside applications.
DeepSeek: record raise lands amid a US federal ban 2 items
DeepSeek reportedly raised about $15B days after US agencies were ordered to remove its products, which matters for teams evaluating open-weight Chinese models.
DeepSeek doubles round to $15 billion, led by CATL and Tencent
DeepSeek reportedly doubled its round to $15B, led by CATL and Tencent.
DHS Orders Agencies to Remove DeepSeek AI from Government Systems
DHS ordered federal agencies to remove DeepSeek AI from government systems.
You are caught up for this edition