Anthropic snaps up Humanloop’s humans

OpenAI restores GPT-4o after rocky GPT-5 debut

After a bumpy GPT-5 rollout, OpenAI is making GPT-4o the default again for paying ChatGPT users—no setting toggles required. Plus, subscribers now get easier access to GPT-4.1, o3, and o4-mini through a new “show additional models” setting.

Why the change: GPT-5’s launch last week came with broken routing that left many users feeling the new model was slower and less capable than GPT-4o. Infrastructure hiccups, inconsistent performance, and the abrupt removal of older models fueled the backlash. By restoring 4o, OpenAI is giving people a smoother way to adjust while bugs are fixed.

New controls and limits: GPT-5’s “Thinking” mode—designed for longer, more complex reasoning tasks—now has a 196K-token window and a 3,000-message weekly cap, after which users can fall back to a lighter “Thinking mini” mode. GPT-4.5 remains Pro-only due to high GPU costs.

Anthropic nabs Humanloop’s founders (and most of the team)

This week, Anthropic hired Humanloop’s three co‑founders—Raza Habib, Peter Hayes, and Jordan Burgess—plus around a dozen engineers/researchers in a classic acqui‑hire. (Anthropic just hired people but didn’t buy assets or IP from Humanloop). 

What Anthropic gets: 

• Humanloop is known for building enterprise tools for prompt management, LLM evaluation, and observability used by customers like Duolingo, Gusto, and Vanta.

• With the company’s talent, Anthropic gets enterprise muscle where it matters: evaluation pipelines, compliance/monitoring, and “does this thing behave safely at scale?” tooling. 

• That dovetails with Anthropic’s push on agentic and coding capabilities and its safety‑first brand—and lines up with features buyers ask for alongside longer context windows and agent workflows.

What this means for Humanloop customers: Humanloop had already told customers it was shutting down ahead of a transaction. If you relied on Humanloop, plan to re‑platform—Anthropic will likely fold similar capabilities into Claude’s enterprise stack rather than revive Humanloop as a standalone.

Google’s Gemini update gets personal

Google is rolling out a new Gemini feature that remembers your past conversations automatically. The update lets the AI recall key details and preferences to tailor responses, like suggesting Japanese food content if you’ve previously discussed a Japan-focused YouTube channel.

Privacy controls are baked in. Memory will be switched on by default, but users can disable it in settings. A new “temporary chats” option will also keep certain conversations out of your history and Google’s model training, with data stored for just 72 hours.

🔊 Calling All Designers Who Care About Precision and Consistency 🔊

Tired of AI tools that churn out the same generic results? Recraft gives designers and teams creative control—without sacrificing speed or quality.

Recraft lets you easily:

• Generate illustrations, logos, and icons with stylistic fidelity

• Design vector assets, product mockups, and branded social posts

• Create mood boards and marketing materials in your brand style

Ready to start creating without compromise?

Make scroll-stopping thumbnails in seconds

Thumbler AI helps YouTube creators design professional-quality thumbnails without the time sink or the design skills. Just enter a text prompt or upload an image, and the tool generates multiple design options you can compare side-by-side with potential video titles. 

How you can use it:

• Boost clicks with AI-optimized designs tailored for YouTube feeds

• Test variations of thumbnails and titles before publishing

• Keep branding consistent across your videos with stored assets

• Save time by automating a typically manual, repetitive task

Pricing: Paid, offers free trial

Automated testing with AI-powered digital workers

Functionize is a cloud-based platform that uses AI to generate, execute, and maintain software tests without heavy coding. Its deep learning models create self-healing tests that adapt when applications change, while NLP lets non-technical users write tests in plain language. 

How you can use it:

• Accelerate QA cycles with automated test creation and updates

• Improve reliability with self-healing tests that adapt to changes

• Enable collaboration between technical and non-technical teams

• Integrate seamlessly with existing CI/CD pipelines and tools

Pricing: Contact sales