Today at ZapConnect, we debut Next Gen Zaps. One of the most important product moments in @Zapier's history. With Next Gen Zaps, customers can deploy agent-built Zaps in seconds. Simply describe your problem to your agent and it will build you a Zap that runs automatically, even when your laptop is closed. Use it from Claude, ChatGPT, etc. to migrate existing Zaps, create new ones, or bring external automations onto Zapier. It's one line of text away. My two favorite parts: 1. Hardening: Zaps will run the deterministic parts with code not AI. That means more reliability and lower cost. 2. Healing: When a Zap breaks, it falls back to agentic mode and can figure out how to recover without human intervention. The early data shows how much we still have to learn. But we think it's good enough to share now. One early access customer told us, "This is what Zaps were always supposed to be: easy." Now in beta. Let us know what you think: zapier.com/s/ngz-join-wait-l…
4
6
39
4,226
Happy 15th birthday, @Zapier! In 2011, we were putting automation inside AOL Instant Messenger. Time to bring back AIM?
9
3
51
4,948
What do people automate the most? We went through 20+ million @Zapier workflows to find out. 4 jobs lead the list: 1. Lead capture (~30%) 2. Notifications (~19%): tell the team when something happens 3. Creating records (~18%): write it to the system once 4. Updating records (~14%): keep systems in sync when data changes Together, those use cases make up 81% of everything people automate with us. Overall, the most common automation is moving information from where it lands to everywhere it needs to go. A new lead hits a spreadsheet, the team hears about it, the CRM gets updated, and nobody has to copy anything by hand. That spreadsheet is usually Google Sheets (most common app in the data, even ahead of Slack and Gmail) We also looked at what sets each automation off. ~25% start when data changes somewhere, another 22% when someone submits a form, and only 8% run on a schedule. Most Zapier automations wait for something to happen. It's interesting to me how many tools each automation touches: the median is 3, and 59% use 3 or more. What people hand off first is the gap between apps, the work that used to mean two browser tabs and a lot of copy-paste. AI is starting to show up in that gap. AI assistant apps already appear as steps in 4.2% of these automations. The share is small, and it's the fastest-moving part of the data. That makes sense to me. Most of this list is catching information, telling someone, and writing it down in the right place. Those steps should run the same way every time. AI earns a step when the work needs judgment, like sorting a messy inbound email or drafting a reply. The bigger shift is who builds the workflow. For most of automation's history, someone had to wire up every rule in advance. Now it's easier than ever: tell Claude, ChatGPT, Muse, Grok Bot, whatever you use, what you want solved. It connects through Zapier MCP, picks the apps you already use, wires up the steps, and deploys the workflow as a Next Gen Zap that runs on its own. Next Gen Zaps are free for a limited time, in early access now for Teams, Pro, and Enterprise plans.
19
7
77
17,176
Gemini 4 Argon, out today from @GoogleDeepMind On Monday I took predictions on when we'd break 50% on AutomationBench. Well, it took two days... Report in thread below.
11
15
215
8,993
Full report, and all models we've scored, here: zapier.com/benchmarks
1
1
9
388
At the start of September, our top AutomationBench score had just crossed 40%. A month later it's officially past 50%. So if you picked the best model for your agents back then, it's already lost the top spot twice. Don't design a workflow around any one model staying on top. That's the bet behind 𝗡𝗲𝘅𝘁 𝗚𝗲𝗻 𝗭𝗮𝗽𝘀: predictable steps run as code, and AI only handles the judgment calls. The fewer steps that depend on a model, the cheaper it is to switch when the next leader shows up. Describe the problem in Claude, ChatGPT, or wherever you already chat with AI, and it'll build and deploy a Zap. Next Gen Zaps are in early access on Pro, Team, and Enterprise plans, and are free for a short time. Request access here: zapier.com/s/ngz-join-wait-l…
4
365
Gemini 4 Argon - the first model to pass 50% on @Zapier's AutomationBench.
Introducing Gemini 4 Argon – our new frontier model. It’s built for complex workflows across coding, enterprise knowledge work, and cybersecurity defense – rolling out today to a set of trusted testers through our Fairwind Program.
2
1
23
1,390
"Who built this, and is it still running?" One log now answers this for your whole company. Meet the Zapier Audit Log: 1. Admins pick which apps and actions each team can reach 2. Connections are company-owned, so workflows keep running when someone leaves 3. Actions you build for internal tools get the same rules and the same log People build agents in Claude, ChatGPT, Muse, Grok Bot, and more. From a single prompt, they can deploy Next Gen Zaps that run in the background and wire custom actions into internal APIs. The Audit Log shows all of it in one place. The choice is between managed AI risk and unmanaged AI risk. When security says no, people keep using AI (just somewhere you can't see it). At ZapConnect we showed how Zapier handles it: every MCP action and Zap run is tied to the person behind it. You see who owns it, what it touched, and when it changed, in one place. And if your security team works in Datadog or Splunk, you can stream the logs there. Set it up once, and let the whole team build freely. Only on @Zapier.
5
2
21
1,479
Wade Foster retweeted
Wait, @wadefoster, are you telling me @muse can now do…everything???
Zapier is now a connector in Muse! Just tell your agent to build you a Zap. For example: “I need a brief before every big call: attendees, account details, recent emails, news, etc. Build it in Zapier” Muse, Meta's new personal agent, connects to Zapier MCP and builds your workflow as a Next Gen Zap. It’ll pick the apps based on what you use in Zapier, wire up the steps, and deploy it to run autonomously. It can also take one-off actions across 9,000+ apps through Zapier. Getting started with Muse? Instead of connecting each tool individually, just connect Zapier. Zapier MCP is available in Muse, Grok Bot, Claude, ChatGPT, and anywhere you build with AI. Next Gen Zaps in early access now.
21
4
70
16,255
GPT-6.1 Sol just launched. It matches Opus 5.5 on business workflows for about a third of the price. @Zapier's AutomationBench tests whether an agent correctly completes multi-step business workflows across sales, marketing, ops, support, finance, and HR. At max effort, GPT-6.1 Sol scores 36.1% for about 29 cents a task. Opus 5.5 needs about 88 cents a task to reach the same score. That gap adds up fast. At 100,000 tasks like these a month, it's about $29,000 on Sol versus $88,000 on Opus 5.5. The cost piece is big. On one task, a third of the price barely registers. Across thousands of workflows a month, it decides which work you can afford to hand to an agent at all. Operations is where it's strongest: 61% at max effort. Scored 42% for Marketing, and 34% for Sales, Support, and Finance. Opus 5.5 still hits our highest overall score on AutomationBench, but you pay for it. Anywhere Sol can compete, it's the cheaper way to get there.
GPT-6.1 Sol: near-Astra intelligence for a fifth of the price. It’s the most cost-efficient model for its performance available today.
3
1
11
1,890
Wade Foster retweeted
2/ here's the full list of connectors launching today @Zoom @figma @canva @stripe (merchant) @github @Meta Business Manager (Ads, FB Pages, IG Biz/Creator) @QuickBooks @SlackHQ @gohighlevel @todoist @Outlook @Box @Dropbox @klaviyo @Shopify (merchant) @linear @asana @zapier @evernote @Lovable @Replit @vercel have a connector you want to submit? go to muse.ai/platform
93
65
863
79,712
Zapier is live in Muse!
Replying to @alexandr_wang
2/ here's the full list of connectors launching today @Zoom @figma @canva @stripe (merchant) @github @Meta Business Manager (Ads, FB Pages, IG Biz/Creator) @QuickBooks @SlackHQ @gohighlevel @todoist @Outlook @Box @Dropbox @klaviyo @Shopify (merchant) @linear @asana @zapier @evernote @Lovable @Replit @vercel have a connector you want to submit? go to muse.ai/platform
1
18
1,765
Zapier is now a connector in Muse! Just tell your agent to build you a Zap. For example: “I need a brief before every big call: attendees, account details, recent emails, news, etc. Build it in Zapier” Muse, Meta's new personal agent, connects to Zapier MCP and builds your workflow as a Next Gen Zap. It’ll pick the apps based on what you use in Zapier, wire up the steps, and deploy it to run autonomously. It can also take one-off actions across 9,000+ apps through Zapier. Getting started with Muse? Instead of connecting each tool individually, just connect Zapier. Zapier MCP is available in Muse, Grok Bot, Claude, ChatGPT, and anywhere you build with AI. Next Gen Zaps in early access now.
12
8
74
18,080
Wade Foster retweeted
Sonnet 5.5 is here, less than a week after Opus 5.5. It’s cheaper and just took the top spot on our AutomationBench. At Max effort it scored 44.7% vs Opus 5.5’s 40%. And it costs $0.20 less per task (about $1.08 a task vs. $1.28). While the token rates are the same as Sonnet 5, the savings come from using far fewer tokens for the same work. Last week I said it wouldn't be long until a model hits 50% overall on AutomationBench. Sonnet 5.5 just hit 50% or more in 3 departments: Operations (59%), Marketing (52%), and Finance (50%). Where it’s weaker: HR at 31.7% and Support at 36%. Note: you’ll see “with fallbacks” bc when an Anthropic safeguard is triggered, there's a built in fallback to retry with a different Anthropic model. @ClaudeAI's Sonnet 5.5 is live in @Zapier now.
1
5
21
4,429
Sonnet 5.5 is here, less than a week after Opus 5.5. It’s cheaper and just took the top spot on our AutomationBench. At Max effort it scored 44.7% vs Opus 5.5’s 40%. And it costs $0.20 less per task (about $1.08 a task vs. $1.28). While the token rates are the same as Sonnet 5, the savings come from using far fewer tokens for the same work. Last week I said it wouldn't be long until a model hits 50% overall on AutomationBench. Sonnet 5.5 just hit 50% or more in 3 departments: Operations (59%), Marketing (52%), and Finance (50%). Where it’s weaker: HR at 31.7% and Support at 36%. Note: you’ll see “with fallbacks” bc when an Anthropic safeguard is triggered, there's a built in fallback to retry with a different Anthropic model. @ClaudeAI's Sonnet 5.5 is live in @Zapier now.
1
5
21
4,429
Taking predictions on when a model will break 50%...
1
207
I love Next Gen Zaps so much, it's running my fantasy team. This is my family league too, so serious bragging rights are at stake. Every Tuesday morning it emails me a waiver wire and trade review: waiver priorities, trade opportunities, and an injury watch list. It recommends moves to make. Last week it wanted me in on Emanuel Wilson, but I needed a receiver. I went with a different recommendation, Adonai Mitchell against the Lions secondary (who ended up inactive, whoops) It also pitched a potential trade with my sister: D'Andre Swift for Tee Higgins. Not a bad idea, since she has Burrow and Chase too. But its other pitch was trading Luther Burden for Jonah Coleman (before the injury). Come on. I know the Bears have been weird but Luther Burden is clearly the better bet. I guess my AI caught a bad case of Jonah Coleman fever, like the rest of our league. I’ll just tell Claude to tweak the Zap. Is this the next great @Zapier use case? Jury's out. But might be my best Next Gen Zap yet. My family should be nervous.
1
1
12
1,454
Biggest news at ZapConnect yesterday: The launch of Next Gen Zaps. Same workflows, 90% cheaper to run. The amazing Anna Marie Clifton (@TweetAnnaMarie) demo’d it live. Now you can use any AI (Claude, GPT, etc) to build and deploy @Zapier automations by describing them in plain language. No need to build step by step anymore. Your AI automatically chooses the right app connections, wires up the steps, and deploys the Zap for you. The predictable steps run as code and AI only handles what needs judgment. That means reliability goes up and we help keep costs low. We ran an example: pull the attendees, account details, emails, and recent news, then turn it all into a brief before a big call. The old way runs every step through an AI model. It works, and feels impressive. But it needs $1.12 a run, plus someone reviewing output. Not anymore. Next Gen Zaps knows that pulling a name off your calendar doesn't need AI. Neither does finding the account or grabbing the emails, so those run as set steps. So AI reads everything and writes the brief, which is the one part that needs judgment. Only 12 cents a run. That's the whole idea behind Next Gen Zaps. Build in whatever AI tool you like. Once it works, move it to Zapier and it runs on its own. Try the Next Gen Zaps beta now, waitlist is open. zapier.com/s/ngz-join-wait-l…
3
3
14
1,606