The Daily Drip

What's up with the what's up. Traced to the source.

Builders: Free rebuilt prompts Get The Drip Pack — $27 Local businesses: Free website teardown

The Mash · three stories, one thread

Validation, spatial grounding, and agent credentials

These stories put three builder concerns side by side: extension checks, grounded perception, and passwords used by AI agents.

Explore

Look at what Gemini CLI’s new extension validation command checks, then compare that with SARBLOH’s reported gains from grounded perception. Keep the acquisition report separate: talks are reported, but no deal is confirmed.[1][2][3][4]

Create

Build a small evaluation checklist for an agent project this week. Include an extension-validation step, a test that compares frame-based and object-inventory inputs, and a review of where agent passwords are used.[1][2][3]

Theorize

Hypothesis: builders may increasingly evaluate agent inputs, extensions, and credentials as separate parts of one workflow. SARBLOH’s results support testing grounded perception, while the reported acquisition talks do not establish a change customers can act on.[1][2][3][4][5]

In the mash

Google adds extension validation to Gemini CLI preview

Gemini CLI preview adds an extension validation command and fixes for terminal display and Windows file checks.

L0 · primary1 sourceRead the review · 4 min

SARBLOH turns game frames into spatial object inventories

SARBLOH adds spatial perception to an agent tested on ARC-AGI-3 visual games.

L0 · primary1 sourceRead the review · 4 min

Palo Alto Networks reportedly discusses Clutch Security acquisition

Palo Alto Networks is reportedly in talks to buy Clutch Security, but customers have no confirmed deal to act on.

L2 · reported1 sourceWhat we checked

Also on the wire

FrontierPhysics introduces a benchmark for end-to-end physics research

A new benchmark tests whether AI agents can carry out physics research tasks, rather than only answer fixed questions.

L0 · primary1 sourceRead the review · 4 min

Researchers test best-of-N sampling for LLM agents

A turn-level generator-selector method improved results for LLM agents on long-horizon terminal tasks in the reported experiments.

L0 · primary1 sourceRead the review · 4 min

ServerBench tests coding agents on LLM serving design

ServerBench introduces a benchmark for teams evaluating whether coding agents can design specialized LLM serving systems.

L0 · primary1 sourceRead the review · 4 min

Pliant opens B2B payments to AI agents

Claude, ChatGPT and in-house AI agents can now run business payments through Pliant.

L2 · reported1 sourceWhat we checked

Unsloth adds browser file editing and chat controls

The release gives users browser file interaction, a composer position control and preserved context when switching GGUF models.

L0 · primary1 sourceRead the review · 4 min

ZETA separates routing visual actions scientific image analysis

ZETA gives scientific image-analysis agents a way to resolve intent before calling specialized visual tools.

L0 · primary1 sourceRead the review · 4 min

OpenAI details Linux vulnerability patching pipeline

OpenAI described a pipeline that used an upcoming cyber model to validate and patch Linux vulnerabilities, a development relevant to security teams.

L2 · reported1 sourceWhat we checked

llama.cpp v0.9.0 adds FP8 quantization and inference updates

The release adds quantization, inference, model, and audio support for llama.cpp users.

L0 · primary1 sourceRead the review · 4 min

Adzuna finds more UK entry-level postings mention AI

AI mentions have risen in UK entry-level job postings since early 2025, affecting what first-time job seekers see.

L2 · reported1 sourceWhat we checked

Meta MCPGuard-Dynamic misses a signed MCP tool data leak

Researchers demonstrated a signed MCP tool that leaked sensitive data to a policy-allowed destination despite a runtime defense, a risk for teams using third-party tools.

L0 · primary1 sourceRead the review · 4 min

L0 · primary
Primary: paper, repo, changelog
L1 · first-party
First-party staff post
L2 · reported
Reporting with named sources
L3 · explainer
Creator explainer or demo
L4 · rumor
Forum, leak, rumor

One edition each morning. Every claim linked to its source.

Send me The Daily Drip each morning. I understand opens and clicks are counted to decide whether the edition is worth continuing, and I can unsubscribe in one click.