What changed in Tech today
Articles, releases and code from Hacker News, Reddit, GitHub and the people building RPA, workflow automation and AI agents — plus what the community pushed to the top today.
-
01
Show HN: Edi Life OS – self-hosted life dashboard with an MCP server for AI
AIEdi Life OS provides an open-source personal dashboard that includes an MCP server, enabling AI agents to directly manage personal habits, tasks, and productivity workflows.
-
02
I built an MCP for After Effects
Redditr/mcp 1h ago
-
03
-
04
TractoHub remote MCP server for proposal-to-project workflows
Reddit -
05
-
06
Why do AI agents fail even when the underlying model is good?
Reddit -
07
The real bottleneck of building with AI isn’t code generation—it’s human review fatigue.
Reddit -
08
MS Forms not deployable via solution
Reddit -
09
Synkinema: MCP video editing on the same timeline as the human editor (MIT)
Reddit -
10
I'm non-technical and built an agent that sources five candidates a day
Reddit -
11
Workplace agent request from boss
Reddit -
12
-
13
[Paper] Stepped MoE: Segment-Level Routing with Configurable Inference Complexity
Reddit -
14
[Paper] DLoop: Looped Speculative Decoding
Reddit -
15
From Investigation Failures to Reliable SOC Agents: Understanding and Improving LLM-Based Alert Triage
AIResearchers introduced AIDA, a multi-agent framework that uses adversarial challenges to reduce LLM false-negative rates in automated alert triage from over 40% to 3.1%.
ArticlearXiv agents 4h ago
-
16
Mental-Models for Multi-Agent Systems
AIA new framework equips multi-agent systems with explicit partner mental models, improving coordination and decision-making by inferring the hidden beliefs and intentions of other agents.
Article -
17
Personalization Matters: Long-Horizon Conversation Agent with User-Centric Information in Online Shopping Interactions
AIResearchers introduced a multi-agent RAG framework for conversational shopping that decomposes tracking, retrieval, and reasoning to maintain consistent user preferences across long, multi-turn interactions.
Article -
18
-
19
ConventionPlay: Capability-Limited Training for Robust Ad-Hoc Collaboration
AIConventionPlay trains AI agents to probe partner capabilities and steer them toward optimal shared conventions, improving ad-hoc collaboration in multi-agent workflows.
Article -
20
Whistle: Speech to Text in 16.9 MB
AIWhistle is an open-source, 16.9 MB speech recognition model that runs locally on CPUs, letting developers add private, fast voice transcription directly to edge automation workflows.
ArticleLobsters 4h ago
-
21
Spatial Pattern Formation from Multi-Agent Learning in Public Goods Dilemmas
AIMulti-agent learning research shows uncoordinated agents create inefficient
Article -
22
Google brings agentic AI to Gemini, starting with businesses
AIGoogle launched an enterprise agentic AI for Gemini that integrates with internal business systems and Model Context Protocol servers to execute multi-step workflows.
ArticleTechCrunch Ent 11h ago
-
23
[AINews] not much happened today
AIAnthropic halved Claude Sonnet 3.5 cache-read pricing, while Google launched a universal Gemini work agent with sub-agent orchestration and persistent memory.
ArticleLatent Space 11h ago
-
24
The model that didn't exist, so you made it yourself
ArticleHugging Face 11h ago
-
25
ttok 0.4
AIThe ttok 0.4 release updates the token-counting CLI utility, adding a model-listing flag and uvx compatibility to help automation scripts manage LLM context limits directly.
ArticleSimon Willison 10h ago
-
26
Kubernetes MCP tools: who owns the job after the agent disconnects?
Reddit -
27
-
28
SMS to Teams for MFA
Reddit -
29
[Paper] EngramEdit: Decoupled Knowledge Updates in LLMs through Conditional Memory
Reddit -
30
My SiteGPT experience, and the alternative I landed on
Reddit -
31
Your agent doesn't need more tools — it needs a job description.
Reddit -
32
-
33
-
34
Technical Book recommendations specifically on AI agent planning, tool calling & orchestration
Reddit -
35
I gave my launch directory an mcp server so agents can look up products without scraping
Reddit -
36
How do you tell whether an MCP tool call actually helped the user?
Reddit -
37
Decentralized collaborative continual learning: A multi-objective minimization-based technique
AIA new multi-objective minimization framework enables decentralized AI agents to learn sequentially across evolving tasks without forgetting previous knowledge, improving collaborative edge automation performance.
Article -
38
-
39
Using Google Sheets as the queue for a lead-enrichment pipeline: the failure modes I hit
Reddit -
40
Built Riviera so you can fix how agents use your API without redeploying your backend
Reddit -
41
Devs who rely heavily on AI coding agents, what still slows you down?
Reddit -
42
Is Anyone Else Seeing the AI Agent Silo Wars? Big Players Locking Us In Their Ecosystems
Reddit -
43
-
44
how do you keep automation from hiding important human replies?
Reddit -
45
CAPTCHA was built to keep bots out. But what happens when we actually want AI agents on the internet?
Reddit -
46
-
47
Why I replaced my 20-step AI agent prompt checklist with a 1976 Unix Makefile
Reddit -
48
-
49
What interesting open-source AI agents do you have?
Reddit -
50
Quantization of Linear-Attention (Qwen & Kimi)
Reddit -
51
Testing agent workflows without connecting real accounts
Reddit -
52
-
53
[showcase] I let ChatGPT inspect my actual Windows PC over MCP
Reddit -
54
StepFun Step 5 preview showing up in openrouter
Reddit -
55
WebMCP: is anyone checking this?
Reddit -
56
-
57
Agent accountability is moving from "later" to "now." Time we worked it out across every agent type.
Reddit -
58
rhp-mcp: make beautiful infographic/editorial style charts using rhp 2.0.
Reddit -
59
-
60
Qwen3.8-27B: 159 tok/s on R9700, 64 tok/s on Strix Halo
Reddit -
61
I gave my AI agent a voice and now I can't live without it!
Reddit