Sapien Weekly Digest

Hello from Denver!

What we’ve been up to:

We went to Eth Denver:
The next DeFi wave is agent to agent transactions. The next infrastructure wave is verification for those transactions.” - Lukas Grapentine

Throughout last week, our Director of Solutions Architecture, Lukas Grapentine, and our Product Manager Ali Malik were in Denver, CO, to speak to our partners to be. ETHDenver 2026 had a clear signal: DeFi is shifting from user driven interactions to agent to agent transactions. The release of OpenClaw accelerated this. Many teams were racing to build agentic trading systems and agent mediated market strategies, and the strongest builders were already thinking beyond trading toward practical workflows like subscription management, price discovery, and identity verification.

At Camp BUIDL we saw the rise of vibe coding in full swing with solo developers spinning up projects in record time. Our goals were simple: see what builders are actually building and connect with teams serious enough to be here in a quieter cycle. We met high quality teams, validated the agent trend, and reconnected with builders who ship.

Reaction to Proof of Quality was strong because the mental model already exists in crypto. Builders understand - “trust but verify.” They also understand that agentic flows magnify failure modes. When autonomous systems trade, route, verify, and execute on each other’s behalf, a missing verification layer becomes an agentic catastrophe amplifier.

Proof of Quality fixes this: In order for these new agentic flows to scale up, it is critical for a verification layer to prevent agentic catastrophe.

The new Task UI:
In case you missed it last week, here’s the long awaited first look at what your new UI will look like once the new app drops:

If you’re a builder, hit us up at [email protected] - We want to hear from you!

🚀 What’s New in the App?

While we’re still in the transition phase, the following tasks are available and taken care of for you!

Current task Overview:

1. 😂Thinking Comic Lines 

2. 😂Thinking Comic Lines QA

3. 👩‍🍳 Gastro Tag – Easy / Intermediate / Hard

4. 👩‍🍳 Gastro Tag QA – Easy / Intermediate / Hard

5. 💬 Emotion Prompt – Easy / Intermediate / Hard

6. 💬 Emotion Prompt QA – Easy / Intermediate / Hard

7. 🧩 Multi Choice Error Review – Easy / Intermediate / Hard

8. 📦 Logic Path Vietnamese QA

What else happened in AI?

The World outside of Sapien:

Anthropic releases Claude Sonnet 4.6:
Anthropic shipped Claude Sonnet 4.6 on February 17, aiming it squarely at production developer workloads, positioning it as the most capable Sonnet model, with upgrades across coding, computer use, long context reasoning, agent planning, knowledge work, and design, plus a 1 million token context window in beta. Sonnet 4.6 will be the default model for Free and Pro users in claude.ai and Claude Cowork, with pricing unchanged from Sonnet 4.5 and starting at $3 per million input tokens and $15 per million output tokens. The release also targets agent reliability, especially for computer use, where Anthropic reports stronger resistance to prompt injection attacks than Sonnet 4.5. Anthropic reports that Claude Code users preferred Sonnet 4.6 over Sonnet 4.5 about 70 percent of the time, citing better context reading and less duplicated logic during edits. On the platform side, the same launch highlights context compaction in beta, which summarizes older conversation turns as the user approaches window limits. Tooling also moves forward, including improved web search and web fetch that can dynamically filter results with code execution before they hit the user’s context budget. Sonnet 4.6 is available across Claude plans and the Claude Developer Platform.

Microsoft announces initiatives to address the global AI divide:
Microsoft outlined new initiatives and investments aimed at expanding AI diffusion and applied deployments across the Global South, announced at the India AI Impact Summit on February 17th. AI capability gaps increasingly reflect infrastructure, skills, and deployment capacity rather than model access; diffusion measurement plus targeted applied programs signal a strategy to operationalize AI benefits beyond early adopter markets.

The company said it is on pace to invest USD 50 billion by decade end to expand AI across the Global South through a five part program spanning infrastructure, skilling, language capabilities, local innovation, and measurement. On infrastructure, Microsoft reported more than USD 8 billion in datacenter investments serving the Global South in its last fiscal year and tied that to a connectivity goal of reaching 250 million unserved or underserved Global South communities, including 100M in Africa, with 117M already reached across Africa via partnerships.

Microsoft says it will increase investments in research and data sharing to track diffusion, building on AI Diffusion Reports and cross country analytic contributions. The applied flagship is a new AI for Good Lab initiative in Kenya that layers AI on satellite data with partners including NASA Harvest and Kenyan institutions to generate timely food security insights.

Google adds music generation to Gemini using Lyria 3:
Google added a beta music generation surface to the Gemini app powered by Google DeepMind’s Lyria 3, producing 30 second tracks from either text prompts or uploaded images and videos. The workflow is intentionally lightweight, making it easy to type a prompt or provide a photo or video, then receive a 30 second track plus auto generated cover art. Under the hood, Lyria 3 extends Google’s prior Lyria line with automatic lyric writing, more control over style, vocals, and tempo, and support for more musically complex compositions.

Provenance is the key architectural choice: every track is embedded with SynthID, Google’s imperceptible watermark for identifying Google AI generated content. Gemini’s verification flow now includes audio, letting a user upload an audio file and ask whether it was generated with Google AI, with the app checking for SynthID and using reasoning to respond. For developers building platforms that will ingest these clips, this combination of watermarking and end user verification provides a clearer moderation and attribution.

Lyria 3 music generation is available to Gemini app users aged 18 and older, launches in English, German, Spanish, French, Hindi, Japanese, Korean, and Portuguese, and offers higher limits for paid subscriber tiers.

More to show off next week!