October 24, 2025 • 13 mins
OpenAI’s new AgentKit dropped, and the internet instantly declared n8n dead. Spoiler: it’s not even close.
In this episode of The AI Breakdown, I put AgentKit and n8n head-to-head after weeks of hands-on testing. I dive into where each tool shines, what they’re actually built for, and which one you should reach for when building your next automation or AI agent.
Mark as Played
Transcript
Episode Transcript
Available transcripts are automatically generated. Complete accuracy is not guaranteed.
(00:00):
Hi everyone.
(00:00):
Andy here and welcome back to the I Breakdown.
As you will have probably heard by now, OpenAI launched Agent Kit earlier this month.
Their new shiny agent builder and the internet quickly declared automation tools like N eight N Dead in the water.
Spoiler alert, they're not.
That said, I'm excited about the possibilities Agent Kit brings, especially as it evolves over time.
(00:24):
So over the past few weeks, I've been getting hands-on with the agent kit building prototype solutions.
Putting it head to head with N eight N, my longtime automation workhorse to see how it stacks up.
I tested both on real world projects, like a prototype that now part automates my weekly AI news roundup.
It hunts down recent stories, filters the best ones, and drafts a script ready for my review.
(00:49):
In today's episode, I'll break down how they compare, where each shines and when to choose one over the other. 12 00:01:00,103.916089099 --> 00:01:04,633.916089099 Agent Kit landed on October 6th at Open AI's Dev Day in San Francisco. 13 00:01:05,323.916089099 --> 00:01:09,913.916089099 It's their new toolkit for building, deploying, and optimizing AI agents. 14 00:01:10,513.916089099 --> 00:01:11,503.916089099 Think of it like Canva. 15 00:01:11,503.916089099 --> 00:01:16,573.916089099 For agents, a visual builder where you can drag, drop, and connect steps without much code. 16 00:01:17,53.916089099 --> 00:01:22,33.916089099 It includes three core components, agent builder, which is the visual canvas itself. 17 00:01:22,393.916089099 --> 00:01:29,203.916089099 Chat kit, a slick way to embed your agents as chat widgets on a website and something called the connector registry. 18 00:01:29,503.916089099 --> 00:01:35,353.916089099 Which lets you connect to tools like Gmail, Google Drive, or Outlook, using the MCP protocol. 19 00:01:35,623.916089099 --> 00:01:42,883.9160891 That's model, context, protocol, if you want the technical term, but philosophically, agent Kit isn't trying to be another workflow engine. 20 00:01:43,333.9160891 --> 00:01:44,503.9160891 It's a reasoning system. 21 00:01:44,833.9160891 --> 00:01:48,973.9160891 It's built for people who want to design behavior more than Connect systems. 22 00:01:49,333.9160891 --> 00:01:53,803.9160891 It's there to help you create and deploy AI centric agents with natural language prompts. 23 00:01:54,583.9160891 --> 00:02:17,473.9160891 While N eight N is a broader, more flexible platform for connecting various apps and services through complex event driven workflows when it comes to scale in his dev day presentation, CEO Sam Altman revealed that chat GPT now serves 800 million weekly active users whilst processing over 6 billion tokens per minute through their API. 24 00:02:17,983.9160891 --> 00:02:21,943.9160891 That's the audience agent kit can potentially tap into and it's still growing. 25 00:02:22,513.9160891 --> 00:02:25,183.9160891 N eight N on the other hand, has been around for years. 26 00:02:25,483.9160891 --> 00:02:31,153.9160891 It's an automation powerhouse with over 1000 integrations, webhooks schedules. 27 00:02:31,483.9160891 --> 00:02:41,263.9160891 You can go no code and use their drag and drop designer, or you can roll your sleeves up and embed coded logic, and it gives you the ability to talk to basically any API. 28 00:02:41,263.9160891 --> 00:02:44,653.9160891 You can imagine it's the Swiss Army knife of automation. 29 00:02:44,833.9160891 --> 00:02:47,203.9160891 It's more technical, but infinitely flexible. 30 00:02:47,623.9160891 --> 00:03:03,133.9160891 With over 151,000 GitHub stars, N eight N ranks among the most popular open source projects out there, and for good reason, it's got a massive community of over 200,000 members building templates and solving real problems together. 31 00:03:04,397.49016422 --> 00:03:07,397.49016422 So how do they compare? Let's start with the agent kit. 32 00:03:07,817.49016422 --> 00:03:09,617.49016422 It's incredibly easy to get going. 33 00:03:10,7.49016422 --> 00:03:15,677.49016422 You can spin up a flow that starts, calls an agent, searches the web and spits out results in minutes. 34 00:03:16,472.49016422 --> 00:03:21,32.49016422 No API keys necessary and no config headaches for simple tasks. 35 00:03:21,32.49016422 --> 00:03:21,872.49016422 It's straightforward. 36 00:03:22,232.49016422 --> 00:03:28,262.49016422 The web search quality is genuinely impressive, much higher fidelity than what I've seen through N eight N. 37 00:03:28,652.49016422 --> 00:03:33,752.49016422 In order to design logic-based workflows, you need to map data to Jason Schemas. 38 00:03:34,232.49016422 --> 00:03:42,272.49016422 Open ai have made this straightforward with a built-in schema designer, accompanied by an embedded AI agent that will help you get up and running quickly. 39 00:03:42,662.49016422 --> 00:03:44,117.49016422 And I really like their widget builder. 40 00:03:45,17.49016422 --> 00:03:49,67.49016422 Lets you create custom UI components using natural language. 41 00:03:49,427.49016422 --> 00:03:54,317.49016422 Then map them to agent outputs, which unlocks richer experiences and great flexibility. 42 00:03:54,947.49016422 --> 00:03:58,727.49016422 They make it dead simple to create something that looks polished outta the box. 43 00:03:59,267.49016422 --> 00:04:01,367.49016422 There are also some nice touches under the hood. 44 00:04:01,667.49016422 --> 00:04:06,257.49016422 Evaluation tools for testing prompts, trace grading, and dataset comparisons. 45 00:04:07,142.49016422 --> 00:04:13,262.49016422 Things that show OpenAI is thinking about how you improve an agent, not just build one companies like RAMP Report. 46 00:04:13,262.49016422 --> 00:04:18,992.49016422 That agent Kit transformed what used to take months of complex orchestration into just a couple of hours. 47 00:04:19,352.49016422 --> 00:04:26,822.49016422 They say the visual canvas keeps product legal and engineering on the same page, slashing iteration cycles by 70%. 48 00:04:27,212.49016422 --> 00:04:36,32.49016422 But when you start to stretch it a bit, one of the first hurdles you'll encounter is the documentation is still light and non-existent for some things. 49 00:04:36,452.49016422 --> 00:04:40,82.49016422 And once you move beyond simple chat flows, it quickly gets tricky. 50 00:04:40,562.49016422 --> 00:04:44,762.49016422 Non-technical users will find this challenging and the UI isn't that intuitive. 51 00:04:45,122.49016422 --> 00:04:50,732.49016422 I'm a veteran techie and I was pulling my hair out trying to do a basic loop to process a set of records. 52 00:04:51,152.49016422 --> 00:04:53,282.49016422 Sounds simple, right? It should have been. 53 00:04:53,492.49016422 --> 00:04:58,892.49016422 And I was quite surprised at how unintuitive their designer can be one of the biggest constraints. 54 00:04:59,162.49016422 --> 00:05:00,572.49016422 And this wasn't unexpected. 55 00:05:00,992.49016422 --> 00:05:03,422.49016422 Agent Kit only runs open AI models. 56 00:05:03,962.49016422 --> 00:05:08,942.49016422 There's no way to mix in Andros, Claude, Google's, Gemini, or local models. 57 00:05:09,212.49016422 --> 00:05:20,222.49016422 So you lose flexibility if you want to combine providers, as I often do in N eight N and in my own Python projects, to pick the right tool for the job, get better control of a cost, and so on. 58 00:05:20,792.49016422 --> 00:05:23,672.49016422 So you're locked into open AI's ecosystem entirely. 59 00:05:24,62.49016422 --> 00:05:28,497.49016422 Another limitation is workflows can only be triggered through chat interactions. 60 00:05:29,132.49016422 --> 00:05:30,752.49016422 You can't set up scheduled jobs. 61 00:05:31,97.49016422 --> 00:05:35,477.49016422 Webhook triggers or event driven automations like you can with N eight N. 62 00:05:36,17.49016422 --> 00:05:40,787.49016422 So if you need something running in the background on a timer, agent kit just isn't designed for that. 63 00:05:41,207.49016422 --> 00:05:42,587.49016422 That said, it's early days. 64 00:05:43,7.49016422 --> 00:05:44,537.49016422 This feels like an MVP. 65 00:05:44,837.49016422 --> 00:05:55,277.49016422 There's a basic UI to build out your workflows and a visual canvas, but you get a strong reasoning foundation as you'd expect from OpenAI to make decisions and process data. 66 00:05:55,652.49016422 --> 00:06:08,492.49016422 You can easily equip your agents with knowledge, for example, with file search and via vector stores, and they have made it simple to build guardrails into your flow so that you can protect against hallucinations, jailbreaks, and other risks. 67 00:06:08,852.49016422 --> 00:06:15,332.49016422 Deployments is built in and you can actually export your workflow as Python code if you want to go into full developer mode. 68 00:06:15,782.49016422 --> 00:06:19,292.49016422 Billing doesn't even start until November 1st, 2025. 69 00:06:19,652.49016422 --> 00:06:23,342.49016422 So open AI is clearly still in the get Feedback and iterate phase. 70 00:06:24,694.65183823 --> 00:06:26,494.65183823 Now let's talk N eight N. 71 00:06:27,34.65183823 --> 00:06:34,834.65183823 I've been using N eight N for a while, and it's just in a different league for automation, but like I say, it serves an entirely different purpose. 72 00:06:35,164.65183823 --> 00:06:43,384.65183823 It's a flexible, powerful, and mature automation tool that can easily connect disparate applications, and you can trigger workflows from just about anything. 73 00:06:43,864.65183823 --> 00:06:48,34.65183823 Emails, slack messages, web hooks, CRMs, calendars, you name it. 74 00:06:48,544.65183823 --> 00:06:50,944.65183823 It has over 1000 native integrations. 75 00:06:51,319.65183823 --> 00:06:55,459.65183823 Plus an HTTP node so you can hit any API in existence. 76 00:06:55,819.65183823 --> 00:07:01,639.65183823 This gives it incredible reach, and one thing that really stands out over agent kit is observability. 77 00:07:02,149.65183823 --> 00:07:06,619.65183823 Every node shows you exactly what's going in and out, so debugging is painless. 78 00:07:06,889.65183823 --> 00:07:12,349.65183823 You can replay runs or specific segments of your workflow without having to start the whole thing over. 79 00:07:12,739.65183823 --> 00:07:20,239.65183823 Making the development process efficient, you can inspect variables to see precisely where things broke down or need optimizing. 80 00:07:20,659.65183823 --> 00:07:23,389.65183823 It's one of those platforms that rewards curiosity. 81 00:07:23,629.65183823 --> 00:07:24,769.65183823 You can always dig deeper. 82 00:07:25,339.65183823 --> 00:07:30,889.65183823 They've also got an AI agent built in now, which tries to create workflows for you from a prompt. 83 00:07:31,369.65183823 --> 00:07:43,399.65183823 It's handy, especially if you're new to the tool though, like most AI assistants, it still can't fix its own mistakes, which can be a bit frustrating, but honestly, the value outweighs the annoyance. 84 00:07:43,759.65183823 --> 00:07:47,779.65183823 Within N eight N, you can actually replicate a lot of what Agent Kit does. 85 00:07:48,169.65183823 --> 00:07:49,669.65183823 It has the key foundations. 86 00:07:49,984.65183823 --> 00:07:58,504.65183823 AI agents, web search, knowledge integration, even chat interfaces and NA 10 gives you a much bigger toolbox overall. 87 00:07:58,864.65183823 --> 00:08:02,74.65183823 But I'll be honest, agent Kits chat kit just looks better. 88 00:08:02,824.65183823 --> 00:08:11,464.65183823 Those builtin widgets and the ability to create your own via prompt, it's properly polished in a way, and it NS chat implementation isn't quite there yet. 89 00:08:11,764.65183823 --> 00:08:19,114.65183823 And while N eight N supports multiple model providers, the reliability of non-open AI models for more complex tool-based agents. 90 00:08:19,189.65183823 --> 00:08:20,719.65183823 It isn't always great yet. 91 00:08:21,49.65183823 --> 00:08:30,919.65183823 So even though you can switch providers like OpenAI, Andro, Gemini, whatever, OpenAI still tends to give the most consistent results for complex agent workflows. 92 00:08:31,759.65183823 --> 00:08:36,319.65183823 Another N eight and weak spot for me is web search quality for deep research. 93 00:08:36,859.65183823 --> 00:08:43,69.65183823 For example, I used perplexity models inside N eight 10 to do the research for my news podcast episodes. 94 00:08:43,249.65183823 --> 00:08:47,389.65183823 And what it gave me was pretty underwhelming compared to agent kit's built in search. 95 00:08:47,749.65183823 --> 00:08:55,429.65183823 So when you need high quality research or in-depth analysis, agent kit's, web search just works better and was a lot easier to set up. 96 00:08:55,819.65183823 --> 00:09:05,599.65183823 Still, N eight N is my go-to for anything serious, especially for automations that run in the background, touch multiple systems, or need a proper order trail. 97 00:09:06,334.65183823 --> 00:09:14,854.65183823 With features like role-based access control, audit logs, and GIT integration for version control, it's genuinely enterprise ready. 98 00:09:16,338.67227212 --> 00:09:16,608.67227212 Right. 99 00:09:16,608.67227212 --> 00:09:19,308.67227212 Let's wrap this up with a quick recap on how they compare. 100 00:09:19,668.67227212 --> 00:09:29,928.67227212 Firstly, when considering speed to first useful result, that one goes to Agent Kit, while at least when focusing on agent centric solutions, it's super beginner friendly. 101 00:09:30,258.67227212 --> 00:09:34,968.67227212 Quick to prototype something simple, and the templates mean you can have something running in minutes. 102 00:09:35,268.67227212 --> 00:09:42,648.67227212 But don't forget that NA 10 does offer its own AI agent to help you spin up workflows quickly handling most of the heavy lifting for you. 103 00:09:42,918.67227212 --> 00:09:44,748.67227212 So in theory, it's not far behind. 104 00:09:45,288.67227212 --> 00:09:56,928.67227212 In fact, this is probably the quicker of the two when building more complex workflows when it comes to options to trigger automations and the ability to run these in the background, that's where n it n shines. 105 00:09:57,543.67227212 --> 00:10:03,453.67227212 They have webhooks scheduled events all ticking away quietly behind the scenes while you get on with your day. 106 00:10:04,113.67227212 --> 00:10:14,73.67227212 And in terms of integrations to connect your apps and systems, N eight N wins again, it's got over 1000 integrations and connectors compared to agent kits. 107 00:10:14,73.67227212 --> 00:10:14,913.67227212 Small library. 108 00:10:15,333.67227212 --> 00:10:22,863.67227212 Basically, if it has an API, you can hook it up in N eight 10 and remember that N eight N supports multiple. 109 00:10:23,208.67227212 --> 00:10:34,908.67227212 AI providers like OpenAI and Thropic and Google Agent Kit is OpenAI only, so full vendor Lockin In terms of user interface, agent kit's, chat kit is modern, sleek, and ready for production. 110 00:10:35,178.67227212 --> 00:10:39,78.67227212 It's genuinely polished and outs shines the N eight and equivalent. 111 00:10:39,588.67227212 --> 00:10:43,68.67227212 What about debugging and control? For me, that's all N eight N. 112 00:10:43,488.67227212 --> 00:10:48,948.67227212 Whilst you get some decent tools with an open ai, N 10 provides much better observability. 113 00:10:49,308.67227212 --> 00:10:50,988.67227212 You can easily track data flow. 114 00:10:51,318.67227212 --> 00:10:54,528.67227212 You've got proper error handling and just better overall control. 115 00:10:55,308.67227212 --> 00:11:00,168.67227212 And N eight N also provides more options when it comes to hosting and data ownership. 116 00:11:00,558.67227212 --> 00:11:05,358.67227212 You can self-host it or use their cloud agent kits all managed by OpenAI. 117 00:11:05,658.67227212 --> 00:11:07,908.67227212 So you're basically trusting them with everything. 118 00:11:08,298.67227212 --> 00:11:10,248.67227212 But this also comes with convenience. 119 00:11:10,548.67227212 --> 00:11:12,78.67227212 And finally, let's look at cost. 120 00:11:12,408.67227212 --> 00:11:18,48.67227212 If you self-host N eight N is free to use with unlimited workflows and executions. 121 00:11:18,438.67227212 --> 00:11:23,58.67227212 You manage your own server updates and support and bear those costs. 122 00:11:23,598.67227212 --> 00:11:30,48.67227212 Their cloud plans start at $20 per month and includes 2,500 executions and basic support. 123 00:11:30,708.67227212 --> 00:11:33,288.67227212 This is suitable for individuals or small teams. 124 00:11:33,603.67227212 --> 00:11:35,223.67227212 And you only pay for full runs. 125 00:11:35,283.67227212 --> 00:11:38,133.67227212 Not every little step, so it can be pretty efficient. 126 00:11:38,613.67227212 --> 00:11:45,243.67227212 But remember, regardless of if you go self-hosted or opt for their cloud service, you will still pay for third party services. 127 00:11:45,513.67227212 --> 00:11:50,553.67227212 And API calls like using an open AI API, or other services like serp. 128 00:11:51,483.67227212 --> 00:11:55,893.67227212 When it comes to agent kit, billing begins November 1st with no charges before that date. 129 00:11:56,343.67227212 --> 00:12:00,183.67227212 Each account includes one gigabyte of free monthly storage for chat kit file. 130 00:12:00,798.67227212 --> 00:12:05,568.67227212 And image uploads after which usage is billed at 10 cents per gigabyte per day. 131 00:12:06,48.67227212 --> 00:12:08,88.67227212 Some components remain completely free. 132 00:12:08,388.67227212 --> 00:12:17,418.67227212 You can design and iterate an agent builder at no cost until you hit run self-host jacket and pay only standard model token charges and access. 133 00:12:17,418.67227212 --> 00:12:23,208.67227212 Enterprise controls like SSO are BAC and audit logs without extra fees. 134 00:12:23,718.67227212 --> 00:12:30,378.67227212 The workflow automation market is massive, and whilst the estimates vary, it's currently valued at about $20 billion. 135 00:12:30,918.67227212 --> 00:12:33,678.67227212 Projected to reach 80 billion by 2030. 136 00:12:34,488.67227212 --> 00:12:45,438.67227212 Agent kit positions open AI to capture a meaningful slice of that, especially for teams already using chat GPT or the open ai API and eight N continues to expand too. 137 00:12:45,918.67227212 --> 00:12:54,588.67227212 Its AI agent features are improving, the communities exploding, and its cementing itself as the open standard for building event driven automations. 138 00:12:54,948.67227212 --> 00:13:01,908.67227212 And because it's open source with a self-hosted option, it appeals to organizations that need full control over their data. 139 00:13:02,688.67227212 --> 00:13:05,658.67227212 The truth is these tools aren't really competitors. 140 00:13:06,48.67227212 --> 00:13:11,28.67227212 They just live on different planets, and they both have the mission to save us time through automation. 141 00:13:11,808.67227212 --> 00:13:16,8.67227212 If you need something that talks to people and looks great doing it, try Agent Kit. 142 00:13:16,548.67227212 --> 00:13:20,178.67227212 It's built for customer facing chat, assistant support bots. 143 00:13:20,538.67227212 --> 00:13:26,328.67227212 Or scenarios where the user experience needs a conversational approach to problem solving querying data. 144 00:13:26,418.67227212 --> 00:13:34,308.67227212 And similar, if you need something that talks to systems, runs quietly in the background and gives you full control stick with NHN. 145 00:13:34,638.67227212 --> 00:13:40,668.67227212 It's built for developers and technical teams who need reliability, flexibility, and transparency. 146 00:13:41,932.9158854 --> 00:13:42,562.9158854 And that's it. 147 00:13:42,892.9158854 --> 00:13:44,782.9158854 Thanks for listening to the AI breakdown. 148 00:13:45,22.9158854 --> 00:13:46,72.9158854 Catch you next time.
Hi everyone.
(00:00):
Andy here and welcome back to the I Breakdown.
As you will have probably heard by now, OpenAI launched Agent Kit earlier this month.
Their new shiny agent builder and the internet quickly declared automation tools like N eight N Dead in the water.
Spoiler alert, they're not.
That said, I'm excited about the possibilities Agent Kit brings, especially as it evolves over time.
(00:24):
So over the past few weeks, I've been getting hands-on with the agent kit building prototype solutions.
Putting it head to head with N eight N, my longtime automation workhorse to see how it stacks up.
I tested both on real world projects, like a prototype that now part automates my weekly AI news roundup.
It hunts down recent stories, filters the best ones, and drafts a script ready for my review.
(00:49):
In today's episode, I'll break down how they compare, where each shines and when to choose one over the other. 12 00:01:00,103.916089099 --> 00:01:04,633.916089099 Agent Kit landed on October 6th at Open AI's Dev Day in San Francisco. 13 00:01:05,323.916089099 --> 00:01:09,913.916089099 It's their new toolkit for building, deploying, and optimizing AI agents. 14 00:01:10,513.916089099 --> 00:01:11,503.916089099 Think of it like Canva. 15 00:01:11,503.916089099 --> 00:01:16,573.916089099 For agents, a visual builder where you can drag, drop, and connect steps without much code. 16 00:01:17,53.916089099 --> 00:01:22,33.916089099 It includes three core components, agent builder, which is the visual canvas itself. 17 00:01:22,393.916089099 --> 00:01:29,203.916089099 Chat kit, a slick way to embed your agents as chat widgets on a website and something called the connector registry. 18 00:01:29,503.916089099 --> 00:01:35,353.916089099 Which lets you connect to tools like Gmail, Google Drive, or Outlook, using the MCP protocol. 19 00:01:35,623.916089099 --> 00:01:42,883.9160891 That's model, context, protocol, if you want the technical term, but philosophically, agent Kit isn't trying to be another workflow engine. 20 00:01:43,333.9160891 --> 00:01:44,503.9160891 It's a reasoning system. 21 00:01:44,833.9160891 --> 00:01:48,973.9160891 It's built for people who want to design behavior more than Connect systems. 22 00:01:49,333.9160891 --> 00:01:53,803.9160891 It's there to help you create and deploy AI centric agents with natural language prompts. 23 00:01:54,583.9160891 --> 00:02:17,473.9160891 While N eight N is a broader, more flexible platform for connecting various apps and services through complex event driven workflows when it comes to scale in his dev day presentation, CEO Sam Altman revealed that chat GPT now serves 800 million weekly active users whilst processing over 6 billion tokens per minute through their API. 24 00:02:17,983.9160891 --> 00:02:21,943.9160891 That's the audience agent kit can potentially tap into and it's still growing. 25 00:02:22,513.9160891 --> 00:02:25,183.9160891 N eight N on the other hand, has been around for years. 26 00:02:25,483.9160891 --> 00:02:31,153.9160891 It's an automation powerhouse with over 1000 integrations, webhooks schedules. 27 00:02:31,483.9160891 --> 00:02:41,263.9160891 You can go no code and use their drag and drop designer, or you can roll your sleeves up and embed coded logic, and it gives you the ability to talk to basically any API. 28 00:02:41,263.9160891 --> 00:02:44,653.9160891 You can imagine it's the Swiss Army knife of automation. 29 00:02:44,833.9160891 --> 00:02:47,203.9160891 It's more technical, but infinitely flexible. 30 00:02:47,623.9160891 --> 00:03:03,133.9160891 With over 151,000 GitHub stars, N eight N ranks among the most popular open source projects out there, and for good reason, it's got a massive community of over 200,000 members building templates and solving real problems together. 31 00:03:04,397.49016422 --> 00:03:07,397.49016422 So how do they compare? Let's start with the agent kit. 32 00:03:07,817.49016422 --> 00:03:09,617.49016422 It's incredibly easy to get going. 33 00:03:10,7.49016422 --> 00:03:15,677.49016422 You can spin up a flow that starts, calls an agent, searches the web and spits out results in minutes. 34 00:03:16,472.49016422 --> 00:03:21,32.49016422 No API keys necessary and no config headaches for simple tasks. 35 00:03:21,32.49016422 --> 00:03:21,872.49016422 It's straightforward. 36 00:03:22,232.49016422 --> 00:03:28,262.49016422 The web search quality is genuinely impressive, much higher fidelity than what I've seen through N eight N. 37 00:03:28,652.49016422 --> 00:03:33,752.49016422 In order to design logic-based workflows, you need to map data to Jason Schemas. 38 00:03:34,232.49016422 --> 00:03:42,272.49016422 Open ai have made this straightforward with a built-in schema designer, accompanied by an embedded AI agent that will help you get up and running quickly. 39 00:03:42,662.49016422 --> 00:03:44,117.49016422 And I really like their widget builder. 40 00:03:45,17.49016422 --> 00:03:49,67.49016422 Lets you create custom UI components using natural language. 41 00:03:49,427.49016422 --> 00:03:54,317.49016422 Then map them to agent outputs, which unlocks richer experiences and great flexibility. 42 00:03:54,947.49016422 --> 00:03:58,727.49016422 They make it dead simple to create something that looks polished outta the box. 43 00:03:59,267.49016422 --> 00:04:01,367.49016422 There are also some nice touches under the hood. 44 00:04:01,667.49016422 --> 00:04:06,257.49016422 Evaluation tools for testing prompts, trace grading, and dataset comparisons. 45 00:04:07,142.49016422 --> 00:04:13,262.49016422 Things that show OpenAI is thinking about how you improve an agent, not just build one companies like RAMP Report. 46 00:04:13,262.49016422 --> 00:04:18,992.49016422 That agent Kit transformed what used to take months of complex orchestration into just a couple of hours. 47 00:04:19,352.49016422 --> 00:04:26,822.49016422 They say the visual canvas keeps product legal and engineering on the same page, slashing iteration cycles by 70%. 48 00:04:27,212.49016422 --> 00:04:36,32.49016422 But when you start to stretch it a bit, one of the first hurdles you'll encounter is the documentation is still light and non-existent for some things. 49 00:04:36,452.49016422 --> 00:04:40,82.49016422 And once you move beyond simple chat flows, it quickly gets tricky. 50 00:04:40,562.49016422 --> 00:04:44,762.49016422 Non-technical users will find this challenging and the UI isn't that intuitive. 51 00:04:45,122.49016422 --> 00:04:50,732.49016422 I'm a veteran techie and I was pulling my hair out trying to do a basic loop to process a set of records. 52 00:04:51,152.49016422 --> 00:04:53,282.49016422 Sounds simple, right? It should have been. 53 00:04:53,492.49016422 --> 00:04:58,892.49016422 And I was quite surprised at how unintuitive their designer can be one of the biggest constraints. 54 00:04:59,162.49016422 --> 00:05:00,572.49016422 And this wasn't unexpected. 55 00:05:00,992.49016422 --> 00:05:03,422.49016422 Agent Kit only runs open AI models. 56 00:05:03,962.49016422 --> 00:05:08,942.49016422 There's no way to mix in Andros, Claude, Google's, Gemini, or local models. 57 00:05:09,212.49016422 --> 00:05:20,222.49016422 So you lose flexibility if you want to combine providers, as I often do in N eight N and in my own Python projects, to pick the right tool for the job, get better control of a cost, and so on. 58 00:05:20,792.49016422 --> 00:05:23,672.49016422 So you're locked into open AI's ecosystem entirely. 59 00:05:24,62.49016422 --> 00:05:28,497.49016422 Another limitation is workflows can only be triggered through chat interactions. 60 00:05:29,132.49016422 --> 00:05:30,752.49016422 You can't set up scheduled jobs. 61 00:05:31,97.49016422 --> 00:05:35,477.49016422 Webhook triggers or event driven automations like you can with N eight N. 62 00:05:36,17.49016422 --> 00:05:40,787.49016422 So if you need something running in the background on a timer, agent kit just isn't designed for that. 63 00:05:41,207.49016422 --> 00:05:42,587.49016422 That said, it's early days. 64 00:05:43,7.49016422 --> 00:05:44,537.49016422 This feels like an MVP. 65 00:05:44,837.49016422 --> 00:05:55,277.49016422 There's a basic UI to build out your workflows and a visual canvas, but you get a strong reasoning foundation as you'd expect from OpenAI to make decisions and process data. 66 00:05:55,652.49016422 --> 00:06:08,492.49016422 You can easily equip your agents with knowledge, for example, with file search and via vector stores, and they have made it simple to build guardrails into your flow so that you can protect against hallucinations, jailbreaks, and other risks. 67 00:06:08,852.49016422 --> 00:06:15,332.49016422 Deployments is built in and you can actually export your workflow as Python code if you want to go into full developer mode. 68 00:06:15,782.49016422 --> 00:06:19,292.49016422 Billing doesn't even start until November 1st, 2025. 69 00:06:19,652.49016422 --> 00:06:23,342.49016422 So open AI is clearly still in the get Feedback and iterate phase. 70 00:06:24,694.65183823 --> 00:06:26,494.65183823 Now let's talk N eight N. 71 00:06:27,34.65183823 --> 00:06:34,834.65183823 I've been using N eight N for a while, and it's just in a different league for automation, but like I say, it serves an entirely different purpose. 72 00:06:35,164.65183823 --> 00:06:43,384.65183823 It's a flexible, powerful, and mature automation tool that can easily connect disparate applications, and you can trigger workflows from just about anything. 73 00:06:43,864.65183823 --> 00:06:48,34.65183823 Emails, slack messages, web hooks, CRMs, calendars, you name it. 74 00:06:48,544.65183823 --> 00:06:50,944.65183823 It has over 1000 native integrations. 75 00:06:51,319.65183823 --> 00:06:55,459.65183823 Plus an HTTP node so you can hit any API in existence. 76 00:06:55,819.65183823 --> 00:07:01,639.65183823 This gives it incredible reach, and one thing that really stands out over agent kit is observability. 77 00:07:02,149.65183823 --> 00:07:06,619.65183823 Every node shows you exactly what's going in and out, so debugging is painless. 78 00:07:06,889.65183823 --> 00:07:12,349.65183823 You can replay runs or specific segments of your workflow without having to start the whole thing over. 79 00:07:12,739.65183823 --> 00:07:20,239.65183823 Making the development process efficient, you can inspect variables to see precisely where things broke down or need optimizing. 80 00:07:20,659.65183823 --> 00:07:23,389.65183823 It's one of those platforms that rewards curiosity. 81 00:07:23,629.65183823 --> 00:07:24,769.65183823 You can always dig deeper. 82 00:07:25,339.65183823 --> 00:07:30,889.65183823 They've also got an AI agent built in now, which tries to create workflows for you from a prompt. 83 00:07:31,369.65183823 --> 00:07:43,399.65183823 It's handy, especially if you're new to the tool though, like most AI assistants, it still can't fix its own mistakes, which can be a bit frustrating, but honestly, the value outweighs the annoyance. 84 00:07:43,759.65183823 --> 00:07:47,779.65183823 Within N eight N, you can actually replicate a lot of what Agent Kit does. 85 00:07:48,169.65183823 --> 00:07:49,669.65183823 It has the key foundations. 86 00:07:49,984.65183823 --> 00:07:58,504.65183823 AI agents, web search, knowledge integration, even chat interfaces and NA 10 gives you a much bigger toolbox overall. 87 00:07:58,864.65183823 --> 00:08:02,74.65183823 But I'll be honest, agent Kits chat kit just looks better. 88 00:08:02,824.65183823 --> 00:08:11,464.65183823 Those builtin widgets and the ability to create your own via prompt, it's properly polished in a way, and it NS chat implementation isn't quite there yet. 89 00:08:11,764.65183823 --> 00:08:19,114.65183823 And while N eight N supports multiple model providers, the reliability of non-open AI models for more complex tool-based agents. 90 00:08:19,189.65183823 --> 00:08:20,719.65183823 It isn't always great yet. 91 00:08:21,49.65183823 --> 00:08:30,919.65183823 So even though you can switch providers like OpenAI, Andro, Gemini, whatever, OpenAI still tends to give the most consistent results for complex agent workflows. 92 00:08:31,759.65183823 --> 00:08:36,319.65183823 Another N eight and weak spot for me is web search quality for deep research. 93 00:08:36,859.65183823 --> 00:08:43,69.65183823 For example, I used perplexity models inside N eight 10 to do the research for my news podcast episodes. 94 00:08:43,249.65183823 --> 00:08:47,389.65183823 And what it gave me was pretty underwhelming compared to agent kit's built in search. 95 00:08:47,749.65183823 --> 00:08:55,429.65183823 So when you need high quality research or in-depth analysis, agent kit's, web search just works better and was a lot easier to set up. 96 00:08:55,819.65183823 --> 00:09:05,599.65183823 Still, N eight N is my go-to for anything serious, especially for automations that run in the background, touch multiple systems, or need a proper order trail. 97 00:09:06,334.65183823 --> 00:09:14,854.65183823 With features like role-based access control, audit logs, and GIT integration for version control, it's genuinely enterprise ready. 98 00:09:16,338.67227212 --> 00:09:16,608.67227212 Right. 99 00:09:16,608.67227212 --> 00:09:19,308.67227212 Let's wrap this up with a quick recap on how they compare. 100 00:09:19,668.67227212 --> 00:09:29,928.67227212 Firstly, when considering speed to first useful result, that one goes to Agent Kit, while at least when focusing on agent centric solutions, it's super beginner friendly. 101 00:09:30,258.67227212 --> 00:09:34,968.67227212 Quick to prototype something simple, and the templates mean you can have something running in minutes. 102 00:09:35,268.67227212 --> 00:09:42,648.67227212 But don't forget that NA 10 does offer its own AI agent to help you spin up workflows quickly handling most of the heavy lifting for you. 103 00:09:42,918.67227212 --> 00:09:44,748.67227212 So in theory, it's not far behind. 104 00:09:45,288.67227212 --> 00:09:56,928.67227212 In fact, this is probably the quicker of the two when building more complex workflows when it comes to options to trigger automations and the ability to run these in the background, that's where n it n shines. 105 00:09:57,543.67227212 --> 00:10:03,453.67227212 They have webhooks scheduled events all ticking away quietly behind the scenes while you get on with your day. 106 00:10:04,113.67227212 --> 00:10:14,73.67227212 And in terms of integrations to connect your apps and systems, N eight N wins again, it's got over 1000 integrations and connectors compared to agent kits. 107 00:10:14,73.67227212 --> 00:10:14,913.67227212 Small library. 108 00:10:15,333.67227212 --> 00:10:22,863.67227212 Basically, if it has an API, you can hook it up in N eight 10 and remember that N eight N supports multiple. 109 00:10:23,208.67227212 --> 00:10:34,908.67227212 AI providers like OpenAI and Thropic and Google Agent Kit is OpenAI only, so full vendor Lockin In terms of user interface, agent kit's, chat kit is modern, sleek, and ready for production. 110 00:10:35,178.67227212 --> 00:10:39,78.67227212 It's genuinely polished and outs shines the N eight and equivalent. 111 00:10:39,588.67227212 --> 00:10:43,68.67227212 What about debugging and control? For me, that's all N eight N. 112 00:10:43,488.67227212 --> 00:10:48,948.67227212 Whilst you get some decent tools with an open ai, N 10 provides much better observability. 113 00:10:49,308.67227212 --> 00:10:50,988.67227212 You can easily track data flow. 114 00:10:51,318.67227212 --> 00:10:54,528.67227212 You've got proper error handling and just better overall control. 115 00:10:55,308.67227212 --> 00:11:00,168.67227212 And N eight N also provides more options when it comes to hosting and data ownership. 116 00:11:00,558.67227212 --> 00:11:05,358.67227212 You can self-host it or use their cloud agent kits all managed by OpenAI. 117 00:11:05,658.67227212 --> 00:11:07,908.67227212 So you're basically trusting them with everything. 118 00:11:08,298.67227212 --> 00:11:10,248.67227212 But this also comes with convenience. 119 00:11:10,548.67227212 --> 00:11:12,78.67227212 And finally, let's look at cost. 120 00:11:12,408.67227212 --> 00:11:18,48.67227212 If you self-host N eight N is free to use with unlimited workflows and executions. 121 00:11:18,438.67227212 --> 00:11:23,58.67227212 You manage your own server updates and support and bear those costs. 122 00:11:23,598.67227212 --> 00:11:30,48.67227212 Their cloud plans start at $20 per month and includes 2,500 executions and basic support. 123 00:11:30,708.67227212 --> 00:11:33,288.67227212 This is suitable for individuals or small teams. 124 00:11:33,603.67227212 --> 00:11:35,223.67227212 And you only pay for full runs. 125 00:11:35,283.67227212 --> 00:11:38,133.67227212 Not every little step, so it can be pretty efficient. 126 00:11:38,613.67227212 --> 00:11:45,243.67227212 But remember, regardless of if you go self-hosted or opt for their cloud service, you will still pay for third party services. 127 00:11:45,513.67227212 --> 00:11:50,553.67227212 And API calls like using an open AI API, or other services like serp. 128 00:11:51,483.67227212 --> 00:11:55,893.67227212 When it comes to agent kit, billing begins November 1st with no charges before that date. 129 00:11:56,343.67227212 --> 00:12:00,183.67227212 Each account includes one gigabyte of free monthly storage for chat kit file. 130 00:12:00,798.67227212 --> 00:12:05,568.67227212 And image uploads after which usage is billed at 10 cents per gigabyte per day. 131 00:12:06,48.67227212 --> 00:12:08,88.67227212 Some components remain completely free. 132 00:12:08,388.67227212 --> 00:12:17,418.67227212 You can design and iterate an agent builder at no cost until you hit run self-host jacket and pay only standard model token charges and access. 133 00:12:17,418.67227212 --> 00:12:23,208.67227212 Enterprise controls like SSO are BAC and audit logs without extra fees. 134 00:12:23,718.67227212 --> 00:12:30,378.67227212 The workflow automation market is massive, and whilst the estimates vary, it's currently valued at about $20 billion. 135 00:12:30,918.67227212 --> 00:12:33,678.67227212 Projected to reach 80 billion by 2030. 136 00:12:34,488.67227212 --> 00:12:45,438.67227212 Agent kit positions open AI to capture a meaningful slice of that, especially for teams already using chat GPT or the open ai API and eight N continues to expand too. 137 00:12:45,918.67227212 --> 00:12:54,588.67227212 Its AI agent features are improving, the communities exploding, and its cementing itself as the open standard for building event driven automations. 138 00:12:54,948.67227212 --> 00:13:01,908.67227212 And because it's open source with a self-hosted option, it appeals to organizations that need full control over their data. 139 00:13:02,688.67227212 --> 00:13:05,658.67227212 The truth is these tools aren't really competitors. 140 00:13:06,48.67227212 --> 00:13:11,28.67227212 They just live on different planets, and they both have the mission to save us time through automation. 141 00:13:11,808.67227212 --> 00:13:16,8.67227212 If you need something that talks to people and looks great doing it, try Agent Kit. 142 00:13:16,548.67227212 --> 00:13:20,178.67227212 It's built for customer facing chat, assistant support bots. 143 00:13:20,538.67227212 --> 00:13:26,328.67227212 Or scenarios where the user experience needs a conversational approach to problem solving querying data. 144 00:13:26,418.67227212 --> 00:13:34,308.67227212 And similar, if you need something that talks to systems, runs quietly in the background and gives you full control stick with NHN. 145 00:13:34,638.67227212 --> 00:13:40,668.67227212 It's built for developers and technical teams who need reliability, flexibility, and transparency. 146 00:13:41,932.9158854 --> 00:13:42,562.9158854 And that's it. 147 00:13:42,892.9158854 --> 00:13:44,782.9158854 Thanks for listening to the AI breakdown. 148 00:13:45,22.9158854 --> 00:13:46,72.9158854 Catch you next time.
Popular Podcasts
Stuff You Should Know
If you've ever wanted to know about champagne, satanism, the Stonewall Uprising, chaos theory, LSD, El Nino, true crime and Rosa Parks, then look no further. Josh and Chuck have you covered.
My Favorite Murder with Karen Kilgariff and Georgia Hardstark
My Favorite Murder is a true crime comedy podcast hosted by Karen Kilgariff and Georgia Hardstark. Each week, Karen and Georgia share compelling true crimes and hometown stories from friends and listeners. Since MFM launched in January of 2016, Karen and Georgia have shared their lifelong interest in true crime and have covered stories of infamous serial killers like the Night Stalker, mysterious cold cases, captivating cults, incredible survivor stories and important events from history like the Tulsa race massacre of 1921. My Favorite Murder is part of the Exactly Right podcast network that provides a platform for bold, creative voices to bring to life provocative, entertaining and relatable stories for audiences everywhere. The Exactly Right roster of podcasts covers a variety of topics including historic true crime, comedic interviews and news, science, pop culture and more. Podcasts on the network include Buried Bones with Kate Winkler Dawson and Paul Holes, That's Messed Up: An SVU Podcast, This Podcast Will Kill You, Bananas and more.