Test app changes without touching your live app
Today's edition looks at a safer place to try app changes and a way to compare the cost of switching tools. You'll also find an update for apps that call AI models, including a change to Hugging Face routing.
In today’s edition
- Lovable gives app drafts their own backends.
- Hugging Face changes how model requests are routed.
Try Lovable changes away from your live app
Lovable Drafts now let you explore changes without touching your live app. That's useful when you want to try a bigger idea before changing what customers see.
- Each draft comes with its own backend.
- You can use drafts to explore designs, interface flows, and features.
What it means Try a draft when you're planning a redesign or feature for a live Lovable app.
Source: Lovable
Hugging Face adds automatic provider fallback
Hugging Face now routes language-model requests through OpenRouter instead of its own router. If a provider fails, the new router retries with another provider without changing your code.
- OpenRouter selects providers using live price, speed, and reliability measurements. In OpenRouter's measurements, its routing cut capacity errors by 46%.
- Requests use the same Hugging Face token and API, but you can no longer use your own provider API key for these language models.
- You can still select a provider or turn providers off. Image, video, and audio generation continue through Hugging Face's existing system.
What it means If your app uses Hugging Face language models, check your provider and billing settings, especially if you use a provider key.
Source: huggingface.co
Quick hits
- California sued OpenAI over alleged inadequate safeguards in agent testing, including the Hugging Face incident; OpenAI says it takes misalignment seriously.
- OpenAI is starting embedded safety assessments with outside researchers, including METR, which plans a limited alignment assessment and public reports.
- Perplexity launched Frontier Research with at least 10,000 GPUs committed to work toward autonomous research and plans to publish models, data, and evaluations.
- Perplexity launched its Monitor API to track topics or page changes on a schedule and send updates to apps, priced at $1 per 1,000 runs or URLs fetched.
- Liquid's d1.2 decision model adds audio input and new answer types, with self-serve API pricing of $0.02 per million input tokens.
- Isomorphic Labs announced $5 billion in funding to support its AI drug-design work.
GitHub repo of the day
last30days-skill: research a topic across sources
- What it does: The maintainers describe a research skill that searches recent discussions and synthesizes a brief. Reddit, Hacker News, Polymarket, and GitHub work without extra configuration; other sources depend on setup.
- How to start: In Claude Code, run /plugin marketplace add mvanhorn/last30days-skill, then /plugin install last30days. Try /last30days followed by a topic you want to research.
AI workflow of the day
How to decide whether switching tools pays off
Shared by Robin Delta (@heyrobinai) · View original post ↗
If you're weighing a new tool, this prompt helps you compare its price with the work and expense of switching. It asks for low, likely, and high scenarios rather than relying on one estimate.
- Replace the placeholders with your current costs, the new tool's costs, switching work, and improvement estimates.
- Paste the completed prompt into your AI assistant.
|
Help me decide whether switching to a new tool pays off.
Current workflow: [YOUR CURRENT HOURS, LABOUR COST, AND FEES]
New tool: [NEW TOOL PRICE AND RECURRING COSTS]
Switching costs: [SETUP, TRAINING, AND STAFF TIME]
Expected improvement: [LOW, LIKELY, AND HIGH ESTIMATES]
Show the monthly benefit, break-even time, and first-year result for each scenario. Calculate the minimum improvement needed to break even within a year. Show cash savings and time saved separately. Recommend whether to switch, run a pilot, or wait.
|