OpenAI just shipped an AI that does your job. Anthropic shipped one that tells you to take a break from it. Same week.
That contrast is the whole story right now. The race isn't just about capability anymore. It's about who owns your workflow and how deep that hook goes. ChatGPT Work runs for hours. Claude Reflect nudges you to log off. Two very different bets.
Meanwhile, Zhipu's SAO quietly makes agentic RL training leaner. The plumbing is getting better.
|
|
In Partnership with
|
·
|
|
|
|
|
|
Today's Author
Lior Alexander. Founder of AlphaSignal and former ML Engineer. Previously built ML systems at Iguazio, Guesty, Enphase, Mila.
|
|
Top News
|
|
|
|
OpenAI ships ChatGPT Work, a GPT-5.6 agent that runs full workflows for hours
|
|
18,679 Likes
|
|
ChatGPT just got a new mode called ChatGPT Work, and it's a big shift. Instead of answering questions, it actually does the work for you. Think of it like hiring an AI that takes a goal, figures out the steps itself, and hands you back a finished doc, slide deck, spreadsheet, or web app. No hand-holding required. Here's what it can actually do: - Pull data from Google Drive, Slack, Salesforce, and other connected apps
- Stay on a complex project for hours, breaking it into steps and completing them on its own
- Output finished sheets, slides, docs, and shareable web apps, not just chat responses
- Schedule tasks to run automatically
It runs on GPT-5.6, which comes in three tiers: Sol (top), Terra (mid), Luna (budget). Codex is now merged into the ChatGPT desktop app on Mac and Windows, so you get Chat, Work, and Codex all in one place. ChatGPT Work is live now for Pro, Enterprise, and Edu plans, rolling out to Plus and Business soon.
|
|
|
|
Presented by Unblocked
|
|
[Webinar] Your agents are still missing context.
|
|
AI is in your engineering workflow. While the token spend shows it, the throughput doesn't. The human is very much still in the loop, and that's a context problem. Join live July 23 (FREE) to learn: - The 8 levels of context maturity: where most teams are stuck and what the ceiling looks like at each stage
- Why more MCPs, rules, and skills provide agents access but not understanding
- How leading engineering teams are using a context engine to make the most of their agents
|
|
|
|
Top News
|
|
|
|
Anthropic adds monthly usage dashboard to Claude with break nudges and quiet hours
|
|
11,205 Likes
|
|
Claude just shipped Reflect, a usage dashboard that shows you exactly how you've been spending time with the AI. Think Spotify Wrapped, but for your AI habits. Here's what you actually see inside it: - Your most active day, peak hour, and total conversations over the selected period
- A breakdown of topics you've discussed, with a percentage assigned to the ones you bring up most often
- Workflow tips, like suggesting you use Projects if Claude notices you keep re-explaining the same context
- Controls to set quiet hours and schedule break nudges after a chosen amount of Claude use
You can look back on activity over the past 1, 3, 6, or 12 months. It skips incognito chats and anything tied to health integrations. To try it: go to Settings > Reflect. Memory must be turned on, and you need to be on web or desktop.
|
|
|
|
Presented by CopilotKit
|
|
Your agent should learn from every interaction.
|
|
Each one is a signal for improvement. CopilotKit is the only product that observes both Agent and User interactions by Continuous Learning from Human Feedback. Made possible via the open AG-UI protocol. Self learning turns every user correction and preference into a more knowledgeable asset for your product.
|
|
|
|
Top News
|
|
|
|
Cognition releases SWE-1.7, a coding model hitting frontier scores at $1.97 per task
|
|
6,589 Likes
|
|
Cognition just dropped SWE-1.7, their most capable coding model yet, and the story here is about squeezing frontier-level performance without frontier-level cost. The model is built on top of Kimi K2.7 and then improved using reinforcement learning (RL) -- basically a training technique where the model learns by trial and error, getting rewarded for writing code you'd actually want to merge. The tricky part? Long RL training runs are notoriously unstable. Two big problems kept killing progress: entropy collapse, where the model stops exploring new ideas and plateaus, and numerical drift between training and inference. Cognition hunted down and fixed both, letting training keep improving well past where earlier runs stalled. - 42.3% on FrontierCode at $1.97 per task -- near Opus 4.8 and GPT-5.5 level
- Runs at 1000 tokens/second via Cerebras
- Trained across 4 datacenters on 3 continents
- Investigates more before editing -- touches more files as a tradeoff
SWE-1.7 is available today in Devin via Web, Desktop, and CLI. You can use it right now inside Devin to tackle longer, real-world coding tasks.
|
|
|
At Alpha Signal, our mission is to build a sharp, engaged community focused on AI, machine learning, and cutting-edge language models, helping over 200,000 developers stay informed and ahead. We're passionate about curating the best in AI, from top research and trending technical blogs to expert insights and tailored job opportunities. We keep you connected to the breakthroughs and discussions that matter, so you can stay in the loop without endless searching. We also work closely with partners who value the future of AI, including employers and advertisers who want to reach an audience as passionate about AI as we are.
Our partnerships are based on shared values of ethics, responsibility, and a commitment to building a better world through technology.Privacy is a priority at Alpha Signal. Our Privacy Policy clearly explains how we collect, store, and use your personal and non-personal information. By using our website, you accept these terms, which you can review on our website. This policy applies across all Alpha Signal pages, outlining your rights and how to contact us if you want to adjust the use of your information. We're based in the United States. By using our site, you agree to be governed by U.S. laws.
|
|
|
Looking to promote your company, product, service, or event to 250,000+ AI developers?
|
|
|
|
|