Philipp Schmid(@_philschmid)

If you are building real-time voice agents with @GoogleDeepMind Gemini Live, you can now trace your ...

Philipp Schmid (@_philschmid) — Thu, 23 Jul 2026 14:33:15 GMT

If you are building real-time voice agents with @GoogleDeepMind Gemini Live, you can now trace your speech-to-speech agent loops directly in @LangChain!

- Speaker callback hooks capture only the exact audio your user heard before an interrupted event.
- Live timeline highlighting user speech in blue, Gemini responses in orange, and custom tool executions (like a 1.29s city weather lookup).
- Usage metadata separates regular text tokens from native Gemini audio tokens so you always know your exact usage footprint.

To try it out. In your Google ADK agent, register LangSmithGoogleADKLivePlugin and done. 

[Video Demo by Caroline di Vittorio]

Your browser does not support the video tag.

💬11🔄8❤️47👀4501📊17

TICKR App: https://t.co/12DpSDOn0D Docs: https://t.co/tpyziKMyxy

Philipp Schmid (@_philschmid) — Wed, 22 Jul 2026 17:51:25 GMT

TICKR App: aistudio.google.com/apps/bundled/t…

Docs: ai.google.dev/gemini-api/doc…

💬1🔄1❤️2👀1069📊1

Gemini 3.6 Flash is officially the new default model in Gemini Managed Agents. Your agents will run ...

Philipp Schmid (@_philschmid) — Wed, 22 Jul 2026 17:51:24 GMT

Gemini 3.6 Flash is officially the new default model in Gemini Managed Agents. Your agents will run on 3.6 Flash automatically with zero code changes. 

You can still route back to 3.5 Flash or use Gemini 3.5 Flash-Lite anytime by setting the target model ID directly in your agent config.

Try out Gemini 3.6 Flash in Managed Agents right now using our interactive TICKR finance multi-agent app (video) in @GoogleAIStudio, link below.

Your browser does not support the video tag.

💬8🔄6❤️139👀9654📊22

Box: https://t.co/RefxNOAI2q

Philipp Schmid (@_philschmid) — Tue, 21 Jul 2026 18:52:49 GMT

Box: x.com/Box/status/207…

Box@Box

.@GoogleDeepMind released Gemini 3.6 Flash and Gemini 3.5 Flash-Lite. We ran evals. Here is how 3.5 Flash-Lite compared to 3.1 Flash-Lite on real enterprise document tasks:

→ Overall: 58% vs 41%
→ Data Analysis: +27pp
→ Retail: +28pp
→ Financial Services: +16pp
→ Consumer Products: +21pp

A clear generational step up on the hard, quantitative work that drives real decisions — and it gets there faster and lighter than the model it replaces, not slower.

💬0🔄0❤️5👀2091📊1

Gemini 3.5 Flash-Lite is the king of today 👑. - 350 output tokens/sec. - $0.09 per task on Artifi...

Philipp Schmid (@_philschmid) — Tue, 21 Jul 2026 18:51:28 GMT

Gemini 3.5 Flash-Lite is the king of today 👑. 

- 350 output tokens/sec.
- $0.09 per task on Artificial Analysis.
- Rolling out in Google Search at the moment.
- Beats Gemini 3 Flash on coding and computer use.
- 3.5 Flash-Lite is $0.30/1M input and $2.50/1M output.
- Jumped from 41% to 58% on Box AI complex document reasoning evals.

This probably brings agentic ai to billions of new people.

💬14🔄10❤️115👀7494📊26

We wrote a Gemini 3.6 Flash & Gemini 3.5 Flash-lite Developer Guide. All you need to know about ...

Philipp Schmid (@_philschmid) — Tue, 21 Jul 2026 15:46:49 GMT

We wrote a Gemini 3.6 Flash & Gemini 3.5 Flash-lite Developer Guide. All you need to know about the new models 👇🏻

https://t.co/IYUEfJc9WH

💬3🔄6❤️61👀3661📊11

Model Guide: https://t.co/IYUEfJc9WH Blog: https://t.co/RSi2eEsDm6 Try in @GoogleAIStudio: https:/...

Philipp Schmid (@_philschmid) — Tue, 21 Jul 2026 15:31:47 GMT

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

Blog: blog.google/innovation-and…

Try in @GoogleAIStudio: aistudio.google.com/prompts/new_ch…

💬1🔄1❤️13👀1154📊3

Today we are shipping Gemini 3.6 Flash and 3.5 Flash-Lite in GA! - 3.6 Flash is ~20% more token-eff...

Philipp Schmid (@_philschmid) — Tue, 21 Jul 2026 15:31:47 GMT

Today we are shipping Gemini 3.6 Flash and 3.5 Flash-Lite in GA!

- 3.6 Flash is ~20% more token-efficient, better and cheaper than 3.5 Flash.
- 3.6 Flash is $1.50/1M input and $7.50/1M output
- Managed Agents are now powered by `gemini-3.6-flash` 
- 3.5 Flash-Lite is $0.30/1M input and $2.50/1M output
- Available Gemini API, @GoogleAIStudio, @GeminiApp, @Google Search, @Android Studio, Google @antigravity & @googlecloud. 

Over the last weeks, 3.6 Flash has been my main model for building and testing managed agents. It fast, efficient and strong orchestrator. 

I am excited about how this shifts the cost-to-performance for agentic use cases. Read model guide 👇🏻

💬10🔄20❤️165👀7259📊30

YT: https://t.co/768egU4sT8 Blog: https://t.co/aeU5emHBIi

Philipp Schmid (@_philschmid) — Mon, 20 Jul 2026 14:00:03 GMT

YT: youtube.com/watch?v=0vphxN…
Blog: philschmid.de/testing-skills

💬1🔄0❤️4👀1665📊2

At the @aiDotEngineer World's Fair, I gave a talk on why vibe-checking agent skills breaks in produc...

Philipp Schmid (@_philschmid) — Mon, 20 Jul 2026 14:00:02 GMT

At the @aiDotEngineer World's Fair, I gave a talk on why vibe-checking agent skills breaks in production and how to build reliable automated evals before your users find the bugs. I talked about:

 🎯 Why vague descriptions cause skill failures and how adding negative test cases prevents keyword hijacking on unrelated prompts 
✂️ Why skill files over 500 lines degrade model reasoning
⚡ How to validate outcomes using simple, millisecond regex assertions across 10 to 20 production prompts. 
🗑️ How to run ablation tests (with vs. without skills) to know when models have caught up and you can retire skills.

Check out the full talk and resources below:

Your browser does not support the video tag.

💬7🔄7❤️61👀7380📊17

https://t.co/v54oLXJXnF

Philipp Schmid (@_philschmid) — Fri, 17 Jul 2026 16:14:29 GMT

philschmid.de/managed-agents…

💬0🔄0❤️4👀1406📊1

Run automated PR triage with Gemini Managed Agents. Give your agent access to GitHub CLI (`gh`) ins...

Philipp Schmid (@_philschmid) — Fri, 17 Jul 2026 16:14:28 GMT

Run automated PR triage with Gemini Managed Agents.  Give your agent access to GitHub CLI (`gh`) inside a sandboxed environment, and keep your credentials out of untrusted code.

How it works:
1. Give the agent a dummy GH_TOKEN to satisfy local checks
2. Gemini API Intercept calls to api github and github com at the network layer
3. Egress proxy swap in your real token (`Bearer` for API, `Basic` for git)
4. Reuse the container across turns to keep gh installed

The agent gets full CLI access, and your real token stays in the control plane.

💬19🔄7❤️66👀5829📊27

Integrated Gemini 3.5 Flash + Interactions API into @SpaceXAI Grok build. 👀

Philipp Schmid (@_philschmid) — Fri, 17 Jul 2026 12:22:15 GMT

Integrated Gemini 3.5 Flash + Interactions API into @SpaceXAI Grok build. 👀

💬19🔄10❤️200👀28727📊41

🫪🫪🫪

Philipp Schmid (@_philschmid) — Fri, 17 Jul 2026 08:18:36 GMT

🫪🫪🫪

💬11🔄0❤️128👀11461📊21

https://t.co/QsZ5bPQNrH

Philipp Schmid (@_philschmid) — Thu, 16 Jul 2026 17:07:05 GMT

ai.google.dev/gemini-api/doc…

💬0🔄0❤️5👀1106📊1

Free tier for Managed Agents in the @GoogleAIStudio Gemini API! + 2 updates to help you build cost-c...

Philipp Schmid (@_philschmid) — Thu, 16 Jul 2026 17:07:04 GMT

Free tier for Managed Agents in the @GoogleAIStudio Gemini API! + 2 updates to help you build cost-controlled, automated agents:

🆓 Free Tier usage via the Gemini API with keys from free tier projects. 
🛡️ max_total_tokens to pause tasks safely and resume.
⏰ Native cron triggers (schedule: "0 9 * * *")

Let me know what you build or what you want to see next!

💬8🔄11❤️113👀7663📊25

https://t.co/xBpR8Dtu5r

Philipp Schmid (@_philschmid) — Thu, 16 Jul 2026 17:07:04 GMT

x.com/GoogleAIStudio…

Google AI Studio@GoogleAIStudio

x.com/i/article/2077…

💬3🔄0❤️6👀2043📊3

In today's harness engineering experiment, i ran an eval comparing two bash execution styles, persis...

Philipp Schmid (@_philschmid) — Wed, 15 Jul 2026 18:03:55 GMT

In today's harness engineering experiment, i ran an eval comparing two bash execution styles, persistent (cd and environment variables survive) and clean stateless.

Persistent mode sounds obvious at first. You change directories once and keep working. But this can break the agent's spatial awareness.

Persistent worked fine when every action stayed inside bash. They broke when the agent used other tools with “paths” tools like Read or Write. After cd app/data in bash it tried to call Write(path="settings.conf").

I also tested whether stripping `- working directory: <path>` from SI would change it. It fixed the spatial confusion of cd + read, but worsen any other eval as the agent has no idea in where currently is.

💬9🔄2❤️29👀5462📊12

Snippet: https://t.co/HShhBr8xNb Docs: https://t.co/QsZ5bPQNrH

Philipp Schmid (@_philschmid) — Tue, 14 Jul 2026 16:15:03 GMT

Snippet: gist.github.com/philschmid/85b…
Docs: ai.google.dev/gemini-api/doc…

💬2🔄0❤️5👀1409📊3

One API call. That is all it takes to build a specialized financial analyst using Gemini Managed Age...

Philipp Schmid (@_philschmid) — Tue, 14 Jul 2026 16:15:02 GMT

One API call. That is all it takes to build a specialized financial analyst using Gemini Managed Agents. Combine remote MCP server with an skill from GitHub in 1 API call to build an autonomous financial analyst creating slides for you.

The agent gets its own Linux sandbox, called yahoofinance mcp for real-time financials and created a 6-slide executive deck with custom SVG charts using zarazhangrui/frontend-slides.

Experimenting with ideas was never easier. Snippet below

💬10🔄5❤️57👀5132📊19

Reading recommendation. If you build with AI or care about where the field is heading, this is well ...

Philipp Schmid (@_philschmid) — Tue, 14 Jul 2026 11:23:53 GMT

Reading recommendation. If you build with AI or care about where the field is heading, this is well worth 10 minutes of your time.

Proud to be part of @demishassabis team

Demis Hassabis@demishassabis

x.com/i/article/2076…

💬8🔄4❤️71👀8125📊13

ICYMI: We published the Gemma 4 Technical Report! You've been running the models for months. Read ...

Philipp Schmid (@_philschmid) — Mon, 13 Jul 2026 14:25:38 GMT

ICYMI: We published the Gemma 4 Technical Report! You've been running the models for months. 

Read how we achieved memory and inference across the family. How a 5:1 local-to-global attention ratio combined with pp-RoPE reduced the KV cache footprint. Or how speculative decoding and Multi-Token Prediction drafters work.

arxiv.org/abs/2607.02770

💬6🔄4❤️94👀6263📊19

You can now get custom `ai(.)studio` URLs for your apps in @GoogleAIStudio. When you publish, pick ...

Philipp Schmid (@_philschmid) — Fri, 10 Jul 2026 15:01:25 GMT

You can now get custom `ai(.)studio` URLs for your apps in @GoogleAIStudio.

When you publish, pick your own subdomain (like `my-billion-dollar-app(.)ai(.)studio`). It's ready to share on the web immediately, and your code and chat history stay private.

Subdomains are globally unique. Don't wait. Deploy now.

Your browser does not support the video tag.

💬9🔄5❤️97👀8290📊18

Big QoL for @GoogleAIStudio you can now import projects directly from @github and sync them back. ...

Philipp Schmid (@_philschmid) — Wed, 08 Jul 2026 16:33:24 GMT

Big QoL for @GoogleAIStudio you can now import projects directly from @github and sync them back. 

Got to ai(.)studio/build -> click on "+" -> Import from Github

Your browser does not support the video tag.

💬12🔄9❤️106👀10526📊24

Blog: https://t.co/VqdeNsHWG1 Documentation: https://t.co/QsZ5bPQNrH

Philipp Schmid (@_philschmid) — Tue, 07 Jul 2026 16:40:03 GMT

Blog:  blog.google/innovation-and…

Documentation: ai.google.dev/gemini-api/doc…

💬1🔄1❤️12👀1809📊4

Today, we're shipping 4 new capabilities for Managed Agents in the @GoogleDeepMind Gemini API! Backg...

Philipp Schmid (@_philschmid) — Tue, 07 Jul 2026 16:40:02 GMT

Today, we're shipping 4 new capabilities for Managed Agents in the @GoogleDeepMind Gemini API! Background Execution, Remote MCP servers, Custom Function Calling and credentials refresh.

⏱️ Run long-running tasks asynchronously on the server (`background: true`)
🔌 Connect agents directly to internal endpoints via remote MCP servers
⚙️ Combine server-side code execution with local custom function calling
🔑 Refresh API tokens across turns without resetting your sandbox state

We're optimistic about making agents true background teammates rather than interactive chatbots. Let me know what you think, feedback has been incredibly helpful so far!

Documentation and Agent Skill below ⤵️

💬14🔄21❤️233👀21475📊49

need a model for ocr or vqa? try gemini 3.5 flash. gemini 3.5 flash is faster, cheaper, and more acc...

Philipp Schmid (@_philschmid) — Mon, 06 Jul 2026 12:23:49 GMT

need a model for ocr or vqa? try gemini 3.5 flash. gemini 3.5 flash is faster, cheaper, and more accurate. 

Details ↓

💬27🔄12❤️408👀38074📊70

https://t.co/bI45FaTPzJ

Philipp Schmid (@_philschmid) — Mon, 06 Jul 2026 12:23:49 GMT

x.com/skalskip92/sta…

SkalskiP@skalskip92

need a model for OCR and VQA? Fable 5 is NOT your best option

Gemini 3.5 Flash is 1.3x faster, 2.9x cheaper, and more accurate

Fable 5 struggles especially on tasks that require counting objects.

↓ cost comparison and interesting failure cases

💬0🔄2❤️13👀3893📊2

Code: https://t.co/aHiNul4ThN

Philipp Schmid (@_philschmid) — Thu, 02 Jul 2026 16:45:20 GMT

Code: ai.google.dev/gemini-api/doc…

💬0🔄1❤️9👀1558📊2

"Make it day time." The lighting shift, the shadows move, the sky changes. Gemini Omni Flash can edi...

Philipp Schmid (@_philschmid) — Thu, 02 Jul 2026 16:45:19 GMT

"Make it day time." The lighting shift, the shadows move, the sky changes. Gemini Omni Flash can edit your videos through conversation. Upload a clip, describe the change, get a new video back.

All it takes is 12 lines and the Interactions API ⬇️

Your browser does not support the video tag.

💬10🔄13❤️122👀10658📊29

The @googlehealth API got a CLI! ghealth is an open-source CLI for your Fitbit Air. You can now pull...

Philipp Schmid (@_philschmid) — Wed, 01 Jul 2026 17:27:53 GMT

The @googlehealth API got a CLI! ghealth is an open-source CLI for your Fitbit Air. You can now pull 40 health metrics from the terminal or pipe them straight into your AI agent.

Pull sleep breakdowns and heart rate data into context. Use `--detail` flag on sleep gives you per-stage data (awake, deep, REM), which is useful for spotting patterns week over week.

```
npx skills add Google-Health-API/google-health-cli
ghealth setup
```

Setup takes a few minutes. Wizard handles GCP project, OAuth, scopes.

💬18🔄36❤️317👀49101📊103

https://t.co/K5xi1QGkrp

Philipp Schmid (@_philschmid) — Wed, 01 Jul 2026 17:27:53 GMT

github.com/Google-Health-…

💬0🔄0❤️11👀2332📊5

Giving two talks today @aiDotEngineer, come say hi 👋 1:30pm · Expo Stage 3 SW: Why Agents Should H...

Philipp Schmid (@_philschmid) — Wed, 01 Jul 2026 16:29:48 GMT

Giving two talks today @aiDotEngineer, come say hi 👋

1:30pm · Expo Stage 3 SW: Why Agents Should Have Their Own Sandbox. Hands-on Interactions API and Managed Agents + Questions.

3:20pm · Room 2005: Don't Ship Skills Without Evals. Covers different types, how to build a lightweight eval harness and best practices. 

See you there!

💬8🔄4❤️59👀7480📊14

npx skills add google-gemini/gemini-skills --skill gemini-omni-flash-api

Philipp Schmid (@_philschmid) — Tue, 30 Jun 2026 20:43:44 GMT

npx skills add google-gemini/gemini-skills --skill gemini-omni-flash-api

Philipp Schmid@_philschmid

We published a skill for Omni Flash so you can bootstrap video editing into your agent:

```
npx skills add google-gemini/gemini-skills --skill gemini-omni-flash-api
```

The skill covers the core workflows:
- Text to video
- Image references to video
- First frame to video
- Conversational video editing

Your browser does not support the video tag.

💬5🔄4❤️51👀5617📊15

I'm excited to see what y'all build now it's in the API. The editing capabilities were made for crea...

Philipp Schmid (@_philschmid) — Tue, 30 Jun 2026 20:40:16 GMT

I'm excited to see what y'all build now it's in the API. The editing capabilities were made for creative pipelines. github.com/google-gemini/…

💬0🔄1❤️31👀2246📊11

We published a skill for Omni Flash so you can bootstrap video editing into your agent: ``` npx ski...

Philipp Schmid (@_philschmid) — Tue, 30 Jun 2026 20:40:15 GMT

We published a skill for Omni Flash so you can bootstrap video editing into your agent:

```
npx skills add google-gemini/gemini-skills --skill gemini-omni-flash-api
```

The skill covers the core workflows:
- Text to video
- Image references to video
- First frame to video
- Conversational video editing

Your browser does not support the video tag.

💬19🔄41❤️665👀54485📊235

Shipping today: Nano Banana 2 Lite (GA) and Gemini Omni Flash API (preview). 🚀 Nano Banana 2 Lite ...

Philipp Schmid (@_philschmid) — Tue, 30 Jun 2026 16:07:42 GMT

Shipping today: Nano Banana 2 Lite (GA) and Gemini Omni Flash API (preview). 🚀

Nano Banana 2 Lite (`gemini-3.1-flash-lite-image`) generates images in ~4 seconds at $0.034. Omni Flash generates and edits video through conversation at $0.10/sec up to 10s videos (same price as Veo 3.1 Fast.). 

Omni Flash (`gemini-omni-flash-preview`) combines Gemini's world knowledge (physics, history, cultural context) and is our first model shipped natively for the Interactions API. You get real multi-turn video editing. "Make it sunset." "Add rain." "Swap the product." 
The API keeps your session context, so each edit builds on the last (up to 3 sequential edits per session).

💬1🔄3❤️120👀7190📊17

Try it: https://t.co/PsHysI3scj. NB2 Lite: https://t.co/iZhmyQmLIF Omni Flash: https://t.co/KrofYsk...

Philipp Schmid (@_philschmid) — Tue, 30 Jun 2026 16:07:42 GMT

Try it: aistudio.google.com/prompts/new_ch…. 
NB2 Lite: ai.google.dev/gemini-api/doc…
Omni Flash: ai.google.dev/gemini-api/doc…
Blog: blog.google/innovation-and…

💬0🔄0❤️4👀1472📊1

👀👀👀

Philipp Schmid (@_philschmid) — Tue, 30 Jun 2026 10:57:01 GMT

👀👀👀

Logan Kilpatrick@OfficialLoganK

⚡️🍌🔮

💬4🔄1❤️95👀9591📊12

At @aiDotEngineer this week. Will talk about agents and evals! You can find me at the @GoogleDeepMin...

Philipp Schmid (@_philschmid) — Mon, 29 Jun 2026 20:49:02 GMT

At @aiDotEngineer this week. Will talk about agents and evals! You can find me at the @GoogleDeepMind both, at my talks or send me a DM!

💬6🔄0❤️73👀4602📊14

Guide: https://t.co/Oa21ny8fTd WFT Is interactions API? https://t.co/xmujukBscg

Philipp Schmid (@_philschmid) — Fri, 26 Jun 2026 15:59:25 GMT

Guide: ai.google.dev/gemini-api/doc… 
WFT Is interactions API? blog.google/innovation-and…

💬0🔄0❤️5👀1193📊1

The Interactions API introduced background=True to handle long-running, async agent tasks that excee...

Philipp Schmid (@_philschmid) — Fri, 26 Jun 2026 15:59:24 GMT

The Interactions API introduced background=True to handle long-running, async agent tasks that exceed standard HTTP timeout limits. We wrote a guide on how to use it, how to reconnect, stream.

💬6🔄2❤️41👀4092📊14

Team is working hard to fix and improve the billing experience on @GoogleAIStudio. This week we fixe...

Philipp Schmid (@_philschmid) — Fri, 26 Jun 2026 08:19:49 GMT

Team is working hard to fix and improve the billing experience on @GoogleAIStudio. This week we fixed:

- No unrestricted API keys anymore.
- Infra updates to prevent Billing Account cap overages
- Spend rate limits for Tiers 1-3

Next up: fixing project caps! 

Keep the feedback coming!

💬17🔄4❤️108👀8505📊31

Guide (correct link): https://t.co/mTokUFY359

Philipp Schmid (@_philschmid) — Thu, 25 Jun 2026 16:17:55 GMT

Guide (correct link): philschmid.de/gemini-android…

💬0🔄1❤️10👀1568📊2

@browserbase https://t.co/QZuJs170gw

Philipp Schmid (@_philschmid) — Thu, 25 Jun 2026 16:17:32 GMT

@browserbasex.com/_philschmid/st…

Philipp Schmid@_philschmid

x.com/i/article/2070…

💬0🔄0❤️4👀2394📊1

Guide: https://t.co/W770Kur2Gs Repository: https://t.co/gsjRff1tJO

Philipp Schmid (@_philschmid) — Thu, 25 Jun 2026 16:07:46 GMT

Guide: philschmid.de/gemini-android…
Repository: github.com/google-gemini/…

💬1🔄0❤️6👀1874📊2

Yesterday we launched computer use in Gemini 3.5 Flash with browser, mobile, and desktop environment...

Philipp Schmid (@_philschmid) — Thu, 25 Jun 2026 16:07:45 GMT

Yesterday we launched computer use in Gemini 3.5 Flash with browser, mobile, and desktop environments. I put together a quickstart for how to control an Android Phone. 

Single script to install emulator from terminal.
Basic agent loop with interactions API using `adb` to control the phone.
Connects to remote devices too (`adb connect <ip>:5555`).
Same pattern works for iOS with, e.g. simctl.

Your browser does not support the video tag.

Philipp Schmid@_philschmid

x.com/i/article/2070…

💬11🔄32❤️340👀29541📊86

https://t.co/ALXlwGhCjP

Philipp Schmid (@_philschmid) — Thu, 25 Jun 2026 16:05:53 GMT

x.com/i/article/2070…

💬4🔄11❤️92👀41979📊40