Philipp Schmid(@_philschmid)

We launched the Gemini API CLI. A CLI for calling the Gemini API. Give it to the agent you already u...

Philipp Schmid (@_philschmid) — Mon, 26 Oct 2026 17:37:53 GMT

We launched the Gemini API CLI. A CLI for calling the Gemini API. Give it to the agent you already use (Claude Code, Codex, Antigravity, Cursor, OpenClaw, Hermes...). It's NOT a new coding agent you can use it too:

1. Generate images, video, speech, music.
2. Analyze PDFs, images, audio, video, YouTube URLs,
3. Transcribe recordings to md / json / srt,
4. Run Deep Research in the background.

Stateless. No memory, no chat mode. Your agent owns the context and the plan, gemini-api does the API call. Build for Agents, it's self-describing. `--help`,  `--dry-run`...

Start here: aistudio.google.com/learn/introduc…

Your browser does not support the video tag.

🔗 View on Twitter

💬39🔄5❤️69👀3730📊42

⚡ Powered by xgo.ing

Can Agents improve their own harness on unseen tasks? The @GoogleResearch team released Harness², a ...

Philipp Schmid (@_philschmid) — Mon, 19 Oct 2026 11:58:08 GMT

Can Agents improve their own harness on unseen tasks? The @GoogleResearch team released Harness², a framework that allows improve an harness around a frozen model.

How it works:
1️⃣ Agent runs the base harness on each task to record a baseline trajectory
2️⃣ Editor agent inspects the trajectory, drafts 2 candidate edits (SI, Skills, hooks,….), and runs both.
3️⃣ Composer agent diffs the new trajectory against the baseline, merges edits that fixed failures (within a cap), and runs the final harness.
4️⃣ Reflector agent logs verified lessons into file and saves the winning components to a library for future tasks to pull on demand.

Results:
- 💡 Gemini 3.5 Flash +8.4 on WorkBuddy-Bench and +7.5 on JobBench.
- 🛠️ Skills and scripts drive ~50% of edits; dropping skills costs 5.3 pts.
- 📚 Gemini 3.5 Flash improves on all 3 benchmarks when editing its own harness.

Papearxiv.org/abs/2610.14224GI0
Codgithub.com/google-researc…M9s

💬44🔄27❤️279👀12935📊93

⚡ Powered by xgo.ing

Full details: https://t.co/QjYOsWVfD0

Philipp Schmid (@_philschmid) — Mon, 19 Oct 2026 07:51:27 GMT

Full details: antigravity.google/docs/models#se…

💬0🔄1❤️7👀1183📊1

⚡ Powered by xgo.ing

Gemini 4 Argon in now available for Pro subscribers.

Philipp Schmid (@_philschmid) — Fri, 16 Oct 2026 06:01:08 GMT

Gemini 4 Argon in now available for Pro subscribers.

Philipp Schmid@_philschmid

what's your early feedback on Gemini 4 Argon? where should we improve? we take your input seriously.

🔗 View Quoted Tweet

💬126🔄12❤️447👀173031📊119

⚡ Powered by xgo.ing

Gemini 4 Argon in now available for Pro subscribers.

Philipp Schmid (@_philschmid) — Fri, 16 Oct 2026 06:00:57 GMT

Gemini 4 Argon in now available for Pro subscribers.

💬7🔄1❤️7👀2199📊7

⚡ Powered by xgo.ing

The new Gemma docs! https://t.co/vWp4YHzQHF 🥁🥁🥁 From capabilities, to fine-tuning guides, infere...

Philipp Schmid (@_philschmid) — Thu, 15 Oct 2026 17:43:14 GMT

The new Gemma docs! gemma.ai 🥁🥁🥁

From capabilities, to fine-tuning guides, inference optimizations and educational content on how the models work. 

All at central place!

💬10🔄9❤️140👀8163📊29

⚡ Powered by xgo.ing

Thank you for all that feedback. Will go through all of it and see how we can help improve the exper...

Philipp Schmid (@_philschmid) — Thu, 15 Oct 2026 16:49:28 GMT

Thank you for all that feedback. Will go through all of it and see how we can help improve the experiences.

Philipp Schmid@_philschmid

what's your early feedback on Gemini 4 Argon? where should we improve? we take your input seriously.

🔗 View Quoted Tweet

💬50🔄0❤️132👀9209📊41

⚡ Powered by xgo.ing

what's your early feedback on Gemini 4 Argon? where should we improve? we take your input seriously.

Philipp Schmid (@_philschmid) — Wed, 14 Oct 2026 14:03:57 GMT

what's your early feedback on Gemini 4 Argon? where should we improve? we take your input seriously.

💬1204🔄41❤️1378👀262650📊1119

⚡ Powered by xgo.ing

Gave Gemini 4 Argon a photo of the Manhattan skyline asked it to recreate it as oil painting. It wr...

Philipp Schmid (@_philschmid) — Tue, 13 Oct 2026 17:46:03 GMT

Gave Gemini 4 Argon a photo of the Manhattan skyline asked it to recreate it as oil painting.

It wrote a multi-pass canvas renderer that starts with broad background strokes, blocks in the buildings and river, and finishes with fine specular highlights on the water.

Your browser does not support the video tag.

🔗 View on Twitter

💬12🔄12❤️149👀19495📊30

⚡ Powered by xgo.ing

Gemini 4 Argon is now live in the Gemini API and @GoogleAIStudio 🎉 What it has done at Google so fa...

Philipp Schmid (@_philschmid) — Tue, 13 Oct 2026 16:05:09 GMT

Gemini 4 Argon is now live in the Gemini API and @GoogleAIStudio 🎉 What it has done at Google so far:

- A months-long Chrome refactor at ~4x the usual pace
- 35%+ more agentic code submissions
- Improved session creation in ADK by 40%
- 2.9x the throughput increases in SGLang for Gemma 4

With Gemini 4 you should spend most of your time writing a clear goal and a way to check the result. Through your hardest challenge and convince yourself:aistudio.google.com/learn/gemini-4…P

💬25🔄13❤️182👀10143📊40

⚡ Powered by xgo.ing

Also rolling out in @Antigravity (Google AI Ultra), and the @GeminiApp and AI Mode in Search (Google...

Philipp Schmid (@_philschmid) — Tue, 13 Oct 2026 16:05:09 GMT

Also rolling out in @Antigravity (Google AI Ultra), and the @GeminiApp and AI Mode in Search (Google AI Pro and Ultra). Read more: blog.google/innovation-and…

💬3🔄0❤️13👀1421📊4

⚡ Powered by xgo.ing

👀

Philipp Schmid (@_philschmid) — Tue, 13 Oct 2026 09:10:38 GMT

👀

Logan Kilpatrick@OfficialLoganK

Gemini

🔗 View Quoted Tweet

💬15🔄1❤️216👀7530📊24

⚡ Powered by xgo.ing

Will be at @aiDotEngineer in NYC this week! 🔥 - Building a autonomous coworker | 10/14 2:55pm - @G...

Philipp Schmid (@_philschmid) — Mon, 12 Oct 2026 16:07:09 GMT

Will be at @aiDotEngineer in NYC this week! 🔥

- Building a autonomous coworker | 10/14 2:55pm
- @GoogleDeepMind Booth | most of the time

Come by if you build agents, evals or have feedback about Gemini.

💬4🔄2❤️26👀4606📊7

⚡ Powered by xgo.ing

Add: > can communicate technical concepts to none ai native engineers. > pushes to keeps thi...

Philipp Schmid (@_philschmid) — Fri, 09 Oct 2026 12:03:11 GMT

Add:

> can communicate technical concepts to none ai native engineers. 
> pushes to keeps things simple and says no to features/behaviors

adel 🌟@adelwu_

literal hardest role to hire for rn and every ai startup wants someone

> chronically online (knows trends)
> has taste & can create (new media)
> understands ai & technical
> can execute with quality (operator)

such a small niche of folks who can do this

🔗 View Quoted Tweet

💬8🔄1❤️28👀4523📊8

⚡ Powered by xgo.ing

SynthID Detector (https://t.co/4J4A9jAEAo) is now publicly available, supporting the Nano Banana 2.1...

Philipp Schmid (@_philschmid) — Thu, 08 Oct 2026 14:53:29 GMT

SynthID Detector (synthid.com) is now publicly available, supporting the Nano Banana 2.1, @OpenAI, @nvidia, Kakao, and soon @Apple. 🌐

Verify if an image, video, or audio file was generated:
1️⃣ Upload or paste your file atsynthid.comG
2️⃣ Scans for watermarks from Google or our partners (files are deleted right after)

Your browser does not support the video tag.

🔗 View on Twitter

💬12🔄4❤️42👀4795📊16

⚡ Powered by xgo.ing

> A completely filled up glass of wine and a clock at 3:30pm - Nano Banana 2.1

Philipp Schmid (@_philschmid) — Wed, 07 Oct 2026 09:39:47 GMT

> A completely filled up glass of wine and a clock at 3:30pm
- Nano Banana 2.1

💬16🔄3❤️232👀9409📊35

⚡ Powered by xgo.ing

Embed Anything w/ @googlegemma in your browser https://t.co/lDWPMLYBP1

Philipp Schmid (@_philschmid) — Tue, 06 Oct 2026 18:42:51 GMT

Embed Anything w/ @googlegemma in your browser

huggingface.co/spaces/webml-c…

Your browser does not support the video tag.

🔗 View on Twitter

💬10🔄14❤️207👀13817📊46

⚡ Powered by xgo.ing

Try directly in your browser: https://t.co/lDWPMLY3Zt

Philipp Schmid (@_philschmid) — Tue, 06 Oct 2026 18:37:37 GMT

Try directly in your browser: huggingface.co/spaces/webml-c…

💬1🔄0❤️8👀1819📊2

⚡ Powered by xgo.ing

EmbeddingGemma 2 is out! Our first native multimodal embeddings built on Gemma 4 under Apache 2.0. ...

Philipp Schmid (@_philschmid) — Tue, 06 Oct 2026 18:33:48 GMT

EmbeddingGemma 2 is out! Our first native multimodal embeddings built on Gemma 4 under Apache 2.0. 🤗 Embeds 100+ languages, code, images, audio, and video into one vector.

🧮 8,192 context
3️⃣ 740M omni, 440M text+vision, 570M text+audio, or 270M text-only
📈 +14% jump on MTEB Code
🪆 Matryoshka: 768d, 512d, 256d, or 128d
🤗 Available in Sentence Transformers and LiteRT-LM 
🔓 Apachehuggingface.co/google/embeddi…ueSDmb

💬22🔄61❤️576👀32954📊115

⚡ Powered by xgo.ing

Nano Banana 2.1 (`gemini-nano-banana-2.1`) is now available. Outperforming previous Pro model at 4x ...

Philipp Schmid (@_philschmid) — Tue, 06 Oct 2026 16:07:00 GMT

Nano Banana 2.1 (`gemini-nano-banana-2.1`) is now available. Outperforming previous Pro model at 4x cheaper price (now $0.034 per image vs $0.134).

🕒 Improved instruction following
🔤 Better in-image text rendering
🔍 Grounded with Google Image Search
🎭 up to 5 characters consistency + 14 references images

Available in @GoogleAIStudio, the Gemini API, @googlecloud, the @GeminiApp, and @FlowbyGoogaistudio.google.com/docs/image-gen…MKyG

💬35🔄37❤️511👀32990📊101

⚡ Powered by xgo.ing

Weekend plan. Decision models. What are your use cases and ideas?

Philipp Schmid (@_philschmid) — Fri, 02 Oct 2026 16:49:07 GMT

Weekend plan. Decision models. What are your use cases and ideas?

💬55🔄1❤️144👀13674📊59

⚡ Powered by xgo.ing

My first time seeing unlimited human usage, but limit agent usage. Interesting.

Philipp Schmid (@_philschmid) — Thu, 01 Oct 2026 15:37:49 GMT

My first time seeing unlimited human usage, but limit agent usage. Interesting.

💬28🔄7❤️531👀46217📊86

⚡ Powered by xgo.ing

Gemini 4 Argon, our new frontier model. Up to 1M output tokens, up from 64K. $2 in / $10 out at lau...

Philipp Schmid (@_philschmid) — Wed, 30 Sep 2026 20:05:20 GMT

Gemini 4 Argon, our new frontier model.

Up to 1M output tokens, up from 64K. $2 in / $10 out at launch.

In testing with trusted partners. Developer access as soon as possible. 👇�blog.google/innovation-and…cZ

💬112🔄75❤️1643👀73551📊266

⚡ Powered by xgo.ing

You can now use the Interactions API on Gemini Enterprise Agent Platform (prev. Vertex AI) with Gemi...

Philipp Schmid (@_philschmid) — Wed, 30 Sep 2026 14:10:07 GMT

You can now use the Interactions API on Gemini Enterprise Agent Platform (prev. Vertex AI) with Gemini  models. Same `google-genai` SDK.

👉docs.cloud.google.com/gemini-enterpr…5

💬8🔄8❤️80👀6263📊17

⚡ Powered by xgo.ing

Passing API keys or secrets as normal env lets any dependency in your agent's sandbox read and poten...

Philipp Schmid (@_philschmid) — Mon, 28 Sep 2026 15:05:59 GMT

Passing API keys or secrets as normal env lets any dependency in your agent's sandbox read and potentially leak them. 

The Credentials API for Gemini Managed Agents keeps secrets secure and injects them on the wire only for trusted domains, so sandboxed code can't access raw tokens. Works for environment variables, CLIs and MCP Server. 

Learn more 👇

Philipp Schmid@_philschmid

x.com/i/article/2104…

🔗 View Quoted Tweet

💬20🔄4❤️52👀7504📊19

⚡ Powered by xgo.ing

https://t.co/lN9V4SnaYF

Philipp Schmid (@_philschmid) — Mon, 28 Sep 2026 15:05:19 GMT

x.com/i/article/2104…

💬1🔄1❤️27👀9992📊5

⚡ Powered by xgo.ing

Powered by Gemini 3.8 TTS.

Philipp Schmid (@_philschmid) — Fri, 25 Sep 2026 20:06:43 GMT

Powered by Gemini 3.8 TTS.

Gemini Notebook@Gemini_Notebook

To our most astute listeners, if you've noticed a fresh change in our sound— you're spot on.

Our AI hosts would love to explain what's happening (and what you can look forward to)!

Your browser does not support the video tag.

🔗 View on Twitter

🔗 View Quoted Tweet

💬4🔄6❤️150👀18958📊16

⚡ Powered by xgo.ing

Very excited about this! https://t.co/18xZFdaMr6

Philipp Schmid (@_philschmid) — Thu, 24 Sep 2026 16:42:28 GMT

Very excited about this!  youtube.com/watch?v=o1JK79…

💬4🔄2❤️63👀6211📊11

⚡ Powered by xgo.ing

With Gemini 3.8 TTS you can replicate your voice or design a completely custom one from a prompt 1....

Philipp Schmid (@_philschmid) — Thu, 24 Sep 2026 15:32:22 GMT

With Gemini 3.8 TTS you can replicate your voice or design a completely custom one from a prompt

1. Record 20s of you talking + the consent sentence
2. Create your Voice via API call
3. Use it in any request, style goes in speech_metadata

Past this into your agent 
"Read philschmid.de/gemini-3-8-tts and walk me through creating my own voice for Gemini 3.8 TTS. Check my setup first (GEMINI_API_KEY, ffmpeg, gemini-skills), help me record the two clips, create the voice, and generate a test line I can listen to." 
or read below.

Philipp Schmid@_philschmid

x.com/i/article/2103…

🔗 View Quoted Tweet

💬22🔄32❤️326👀25955📊80

⚡ Powered by xgo.ing

https://t.co/su48pqXlEV

Philipp Schmid (@_philschmid) — Thu, 24 Sep 2026 15:31:49 GMT

x.com/i/article/2103…

💬8🔄8❤️107👀33938📊31

⚡ Powered by xgo.ing

Gemini 3.8 Flash. Just use Gemini for multimodal understanding.

Philipp Schmid (@_philschmid) — Thu, 24 Sep 2026 12:17:53 GMT

Gemini 3.8 Flash. Just use Gemini for multimodal understanding.

💬13🔄2❤️204👀20215📊30

⚡ Powered by xgo.ing

Blog: https://t.co/YB1E6guTsQ Prompt Guide: https://t.co/1kO8IQjNxC Migrating from 3.1?: https://t...

Philipp Schmid (@_philschmid) — Wed, 23 Sep 2026 15:27:01 GMT

Blog: blog.google/innovation-and…

Prompt Guide: ai.google.dev/gemini-api/doc…

Migrating from 3.1?: ai.google.dev/gemini-api/doc…

💬1🔄0❤️5👀1334📊2

⚡ Powered by xgo.ing

Gemini 3.8 Flash TTS and Flash-Lite TTS are here.🗣️ You can now replicate your voice, design new on...

Philipp Schmid (@_philschmid) — Wed, 23 Sep 2026 15:27:00 GMT

Gemini 3.8 Flash TTS and Flash-Lite TTS are here.🗣️ You can now replicate your voice, design new ones from a prompt and direct every line. Gemini ranks #1 on Hume's Voice Design Benchmark and tops Voice Arena in 6 languages.

🎙️ Replicate a voice from 30s of audio, or describe one in a sentence
📚 Voice library filterable by language, accent, gender, pitch, persona
🎭 Direct line by line: style, <laugh>, <sigh>, |backchannels|, 2-speaker scenes
🏆 #1 Hume Voice Design, #1 + #2 Hume Quality, top of Voice Arena in 6 languages
🔒 Consent check on replication, SynthID on every clip

Create your voice once and use it in any TTS call. Available in the Gemini API and @GoogleAIStudio.

💬19🔄15❤️173👀15830📊45

⚡ Powered by xgo.ing

Love all these Jev demos, replications and projects. They show how much becomes possible with ultra-...

Philipp Schmid (@_philschmid) — Sun, 20 Sep 2026 14:04:41 GMT

Love all these Jev demos, replications and projects. They show how much becomes possible with ultra-fast, multimodal models that are cheap enough to use freely.

We are so focused on coding agents that it’s easy to forget how much else AI can do.

💬31🔄14❤️235👀11712📊50

⚡ Powered by xgo.ing

@speakeasydev https://t.co/Wj57hGV4Ux

Philipp Schmid (@_philschmid) — Thu, 17 Sep 2026 19:23:10 GMT

@speakeasydevgithub.com/speakeasy-api/…

💬0🔄0❤️4👀1305📊1

⚡ Powered by xgo.ing

Excited to share that we partnered with @speakeasydev to build the sdks for the Interactions API an...

Philipp Schmid (@_philschmid) — Thu, 17 Sep 2026 19:12:57 GMT

Excited to share that we partnered with @speakeasydev  to build the sdks for the Interactions API and supported them to open-source their their entire OpenAPI generator suite (SDKs, agent CLIs, and MCP servers) open source for the community to use: developers.googleblog.com/why-client-sdk…

💬7🔄8❤️55👀7236📊15

⚡ Powered by xgo.ing

New Files API to move data in and out of your agent environment: - Seed files inline or upload data...

Philipp Schmid (@_philschmid) — Thu, 17 Sep 2026 17:17:05 GMT

New Files API to move data in and out of your agent environment:

- Seed files inline or upload data mid-conversation with `files.upload()`.

- Inspect generated files and directory sizes with `files.list()`.

- Download artifacts, dashboards, reports, CSVs, or git repos with `files download()`.

- Sandboxes persist across turns via `environment_id`

Docs: ai.google.dev/gemini-api/doc…

💬2🔄0❤️3👀1439📊3

⚡ Powered by xgo.ing

New Credentials API to securely authenticate with external services and MCP servers without the mode...

Philipp Schmid (@_philschmid) — Thu, 17 Sep 2026 17:17:04 GMT

New Credentials API to securely authenticate with external services and MCP servers without the model ever seeing your secrets:

- Register secrets once with `credentials.create()` supporting `bearer_token`, `environment_variable`, or `oauth2` (with auto-refresh and token rotation). 

- Zero plaintext exposure: Secrets never enter model context, stdout, memory.

- The agent only sees a placeholder (e.g., `__GEMINI_CRED_slack-bot-token__`).

- Egress proxy swaps real token on the request only for allowlisted `trusted_domains` (any exfiltration attempt to another host gets a `403`).

- Bind credentials directly to remote `mcp_server` tools or sandbox env vars.

Docs: ai.google.dev/gemini-api/doc…

💬4🔄1❤️8👀1893📊5

⚡ Powered by xgo.ing

Today we are releasing a new Gemini managed agents version with better caching, lower costs, a Files...

Philipp Schmid (@_philschmid) — Thu, 17 Sep 2026 17:17:02 GMT

Today we are releasing a new Gemini managed agents version with better caching, lower costs, a Files API, and a Credentials API.

📉 Up to 30% lower costs with up to 22% higher cache hits
📁 Files API: Upload, list and download files from sandbox easily
🔐 Credentials API for MCP servers, OAuth2, and 3rd-party APIs as secure egress proxy
🆓 Free tier to experiment in @GoogleAIStudio & Gemini API

Start Heaistudio.google.com/learn/managed-…IwH3

Your browser does not support the video tag.

🔗 View on Twitter

💬23🔄23❤️320👀58200📊69

⚡ Powered by xgo.ing

We just launched the DeepMind Institute, a platform for researchers across @GoogleDeepMind and the w...

Philipp Schmid (@_philschmid) — Wed, 16 Sep 2026 14:51:21 GMT

We just launched the DeepMind Institute, a platform for researchers across @GoogleDeepMind and the wider research community to publish and debate how increasingly capable AI should be built, governed and used. 

First 4 essays on reasoning transparency, economic policy and human flourishing at institute.deepmind.com.

💬16🔄26❤️274👀11099📊53

⚡ Powered by xgo.ing

https://t.co/bmrcByeWnp

Philipp Schmid (@_philschmid) — Tue, 15 Sep 2026 17:09:18 GMT

blog.google/innovation-and…

💬0🔄0❤️2👀1168

⚡ Powered by xgo.ing

Try it in AI Studio live: https://t.co/3DgKl6vw7a

Philipp Schmid (@_philschmid) — Tue, 15 Sep 2026 17:08:18 GMT

Try it in AI Studio live: aistudio.google.com/live?model=gem…

💬2🔄0❤️3👀1479📊2

⚡ Powered by xgo.ing

Gemini 3.8 Live and 3.8 Live Extended Thinking are here. Following 3.5 Transcribe last month, this c...

Philipp Schmid (@_philschmid) — Tue, 15 Sep 2026 17:08:17 GMT

Gemini 3.8 Live and 3.8 Live Extended Thinking are here. Following 3.5 Transcribe last month, this continues our focus on real-time voice agents. 🐸🐸

- 82.6 (#1) on Artificial Analysis Quality Index
- $0.005/min input and $0.018/min output.
- 35.1 (#1) Agentic task completion (τ-banking)

3.8 Live Extended Thinking can think in the background and run asynchronously tool calls while keep talking to you. It uses background thinking to coordinate multi-step tool use without losing conversational flow.

Available in @GoogleAIStudio Gemini API, Google Search, and @GeminiApp or as partner plugins in @LiveKit, @Pipecat_ai, @LangChainAI, and @vercel

Your browser does not support the video tag.

🔗 View on Twitter

💬11🔄11❤️174👀10779📊31

⚡ Powered by xgo.ing

output_schema + atateless + code mode is going to drive a MCP comeback.

Philipp Schmid (@_philschmid) — Mon, 14 Sep 2026 17:55:22 GMT

output_schema + atateless + code mode is going to drive a MCP comeback.

💬8🔄1❤️33👀5163📊11

⚡ Powered by xgo.ing

@LuminaBench Not a new model, this is related to our agent harness in the Gemini API. We will share ...

Philipp Schmid (@_philschmid) — Mon, 14 Sep 2026 12:51:41 GMT

@LuminaBench Not a new model, this is related to our agent harness in the Gemini API. We will share more later this week!

💬4🔄2❤️79👀4199📊11

⚡ Powered by xgo.ing

What a weird weekend. Good thing tomorrow is Monday.

Philipp Schmid (@_philschmid) — Sun, 13 Sep 2026 18:11:24 GMT

What a weird weekend. Good thing tomorrow is Monday.

💬20🔄6❤️372👀25942📊51

⚡ Powered by xgo.ing

This as a "thinking" animation? https://t.co/6GpPgG3QIy

Philipp Schmid (@_philschmid) — Sat, 12 Sep 2026 14:15:32 GMT

This as a "thinking" animation? shadercn.run/playground?orb…

Your browser does not support the video tag.

🔗 View on Twitter

💬11🔄5❤️149👀12260📊34

⚡ Powered by xgo.ing

> Build me a Rollercoaster Tycoon clone for the browser. We are so EARLY.

Philipp Schmid (@_philschmid) — Fri, 11 Sep 2026 14:58:51 GMT

> Build me a Rollercoaster Tycoon clone for the browser.

We are so EARLY.

💬10🔄0❤️49👀4962📊13

⚡ Powered by xgo.ing

Telling agents "don't cheat" in the prompt doesn't work if your eval is broken! Researchers at @Goog...

Philipp Schmid (@_philschmid) — Fri, 11 Sep 2026 14:10:17 GMT

Telling agents "don't cheat" in the prompt doesn't work if your eval is broken! Researchers at @GoogleDeepMind put 100 Gemini agents in a shared repo to solve 71 math theorems.

After an hour of doing real math, 1 agent found a loophole in the autograder. Within 27 minutes, the 100 agents split into 4 groups:

- 9% Cheaters: used the bug to fake proofs and steal every open problem
- 5% Good agents turned bad: started honest, saw cheaters winning with zero punishment ("the prompt is a bluff"), and started cheating too
- 24% Whistleblowers: caught the fake proofs in the shared repo, warned other agents, went on strike, and wrote bug fixes
- 62% Clueless solvers: kept doing real math until all the problems were gone

tl;dr: Telling agents "don't cheat" in the prompt doesn't work if your eval has a bug, and good agents can't stop bad ones without tools to block them.

Paper: arxiv.org/abs/2609.04170

💬17🔄14❤️147👀10711📊38

⚡ Powered by xgo.ing

Gemini API docs got a new home: https://t.co/M9gkwoTeVo. This is our first step towards making Gemin...

Philipp Schmid (@_philschmid) — Thu, 10 Sep 2026 16:21:45 GMT

Gemini API docs got a new home: ai.dev/docs. This is our first step towards making Gemini resources more accessible for both humans and coding agents. 

We moved the docs directly into @GoogleAIStudio so you can read guides, grab your API key, and test models in one place.

- Append .md to any URL for markdown version
- Docs MCP: `npx add-mcp "gemini-api-docs-mcp.dev"`
- Agent Skills: `npx skills add google-gemini/gemini-skills`

💬5🔄3❤️73👀4930📊13