Introducing Merge’s Universal Context Layer.

It substantially decreases token cost while increasing accuracy.

Every company uses AI wrong:

Say your designer makes a deck using Claude. They teach Claude your design systems and typography.

Now a Marketing hire makes a deck. your designer taught Claude gets passed over… why?

Merge fixes this:

All knowledge and memory across your entire company gets stored inside our context layer.

When someone on the org teaches Al, that skill lives across the entire organization instead of staying locked in their session.

Your AI runs faster, cheaper, and with more accuracy. You don’t have to switch systems, incur more costs, or retrain anyone on your team, Merge runs invisibly in the background.

Book a demo here: merge.dev/merge-for-work…

ok 1password killed it with the personal agents integrations

it's so good

👀 👀

x.com/merge_api/stat…

We tested Merge's MCP connectors against the public MCP servers Google provides for Gmail, Calendar and Drive.

Merge gave agents 122 tools across the three, and Google's servers gave them 27.

Agents on Merge also hit tool errors in 15 runs (we're fixing!), compared with 28 on Google's.

Next Tuesday and Wednesday we'll be at ODSC (@_odsc) AI West in San Francisco, booth #23. Come by if you're connecting agents to the tools your customers already use.

Event details: odsc.ai/west/

I heard from @shensi that public MCP servers fail about 40% of the time.

When Merge benchmarked its tools against public MCP servers, it got 99.999% success versus roughly 60%. Most public servers are side projects nobody maintains. That's also why AI security hasn't had its big break before any data can leak.

00:00 Intro
01:42 Building Merge six years before AI
03:24 Rebuilding for agents on nights and weekends
04:53 How agents pick the right tool out of thousands
06:13 Agents only see what's on the menu
06:42 "It's irresponsible to be using AI without a DLP layer"
07:39 Merge for Workforce and model routing
09:16 Why most AI spend is in coding agents, not product
10:45 Why you can't vibe code critical infrastructure
12:00 Public MCP servers succeed 60% of the time
12:59 Is MCP the right protocol?
16:20 Why founders have to be in the discourse
18:30 "If you as the founder are not AI-pilled… that's the ceiling"
21:05 996, experience, and AI-pilled juniors
37:48 "Humans will become the blocker" to the singularity
39:04 The next 10 years of Merge

Watch the full episode of Navigators on YouTube, Spotify, and Apple Podcasts.

You know you’re getting old when you start to like South Bay

Come say hi to @merge_api at AGNTCon + MCPCon by @AgenticAIFdn!

Just onboarded one of the fastest-growing AI companies in the world onto @merge_api.

We've made a lot of product bets over the past year, and it's incredible to see our team's hard work continue to pay off.

✨️

We are at AGNTCon + MCPCon by @AgenticAIFdn!

Stop by booth 44 for a chance to win a Mac Mini and our new hats!

Testing out the new @merge_api SF office before we move in officially next week!

✨️

I've tried just about every AI agent out there, including Muse. Grok Bot stands above the rest. It's running my weekly content publishing, building amazing websites, and even creating invoices in QuickBooks. Stuff I didn't think was possible to automate. @elonmusk @grok @xai @bot

Dots from @OpenAI can now use your company's approved tools through Merge for Workforce.

Same identity, access, and audit controls IT already set for every other agent, and they hold even after an employee closes the app.

merge.dev/blog/bring-dot…

Next week, Oct 28th, @merge_api is teaming up with @foundherhouse and Harvard Undergraduate Women in Entrepreneurship for a fall happy hour at Russell House Tavern.

If you're around Boston, come say hi!

RSVP: luma.com/merge-ytdz

Two frontier models a few points apart on the Intelligence Index can differ 7x in the tokens they burn per task.

GPT-6 Astra (@OpenAI) scores 52.7 and finishes an average Artificial Analysis Intelligence Index task in 27K output tokens. Claude Sonnet 5.5 scores 56.0 and uses in at 38K, while Muse Spark 1.3, Gemini 4 Argon and MiMo-V2.6-Pro all land between 60K and 64K.

Output tokens are what you pay for. Sending simple requests to a leaner model is one of the easiest ways to cut inference spend, and Merge Gateway routes them for you.

Scores via artificialanalysis.ai, week of Oct 5.

The two top models on the Intelligence Index take about twice as long per task: Claude Opus 5.5 and Sonnet 5.5 (@AnthropicAI) vs. Muse Wonder (@Meta).

Muse Wonder finishes an average task in 6.8 minutes with a 52.4 score. Opus 5.5 takes 13.2 and Sonnet 5.5 takes 15.3, scoring rk 1.3 and GPT-6 Astra (@OpenAI) follow at 8.1.

When someone is waiting on the answer, minutes can matter more than a few index points. Merge Gateway lets you send latency-sensitive requests to the fast model and the hard ones to the strong one, behind one endpoint.

Scores via artificialanalysis.ai, week of Oct 12.

New in Merge Unified: Auto-link. Create a linked account directly from credentials you already hold, no reauthorization needed.

Learn more: merge.dev/blog/unified-b…

The top six open-weights models all come from Chinese labs, led by MiMo-V2.6-Pro (@Xiaomi) at 46.3.

DeepSeek's new V4.1 Flash 1017 is right behind at 45.1, then Z AI's GLM-5.3 (44.8), StepFun's Step 5 Preview (43.7), and Kimi K3 (43.6). Meta's Muse Spark Open (@Meta) lands Open weights mean you can run them on your own infrastructure. Merge Gateway routes to self-hosted models alongside hosted ones, behind one endpoint.

Scores via artificialanalysis.ai, week of Oct 19.

MERGE IS HIRING

i've been at merge for almost 4 years, and it has been a wild ride.

we're hiring in sf, nyc and berlin!

check out the roles here: merge.dev/careers

hosted a cute dinner in dallas tonight!