Unlocking New Possibilities With North Cohere’s AI Agents  ## Metadata - Author: [[Hard Fork AI]] - Full Title: Unlocking New Possibilities With North Cohere’s AI Agents - Category: #podcasts - URL: https://share.snipd.com/episode/9b2e4fdb-129b-4b83-a6b4-d8f0ae89fc71 ## Document Notes ## Highlights - **Agents Replace Tasks First** - Agents will first replace discrete tasks rather than whole employee roles. - We should expect gradual task-by-task automation before fully autonomous agent workers arrive. Transcript: Jaeden Schafer So the first thing I want to say is obviously these AI agent, like these tool platforms, the idea is to automate your workflows, automate things that you're doing. That's basically what they're doing. I think there's like this concept with agents that at some point we're gonna be like, hey, you're a marketing agent, like make my, you know, make my sales go crazy. And you know, the agent runs away and pulls off a miraculous miracle. I think less and less, this is where people are believing these agents are going to be. And more and more, it's going to be basically giving it tasks to accomplish. You figured out how to write the perfect newsletter and send it and the whole workflow that goes into that. And so you basically train the agent, make his automation that can go do that task for you. It's kind of interesting because there's two concepts with agents. One is that they're like an employee. I think right now we're like, we'll get there, I think. But right now we're not there. I think right now we're at the, a task. So if an employee does 10 different tasks, you would take one of those tasks and give it to an agent. That's kind of what we're calling agents now. I think in the future, it's going to be the agents, like an employee that can do all your things you train it to do. ([Time 0:01:45](https://share.snipd.com/snip/4d138ea0-0e04-4d14-b95e-e27f350c06fc)) - **Security Is The Adoption Bottleneck** - Enterprise concern about agent security is a major adoption barrier. - Cohere's North emphasizes private deployment so models access company data only behind customer firewalls. Transcript: Jaeden Schafer What I will say is basically one of the big problems with agents that companies have is that they are not very secure. You're giving it your private data and people don't really know where that all leads or goes. There's been a whole bunch of, you know, leaks, things like that. And so Cohere is basically trying to address that concern and keep their customers' data private. Who are they working with? We're working with big banks, RBC Bank, and a bunch of other like financial places and those kind of places. A lot of enterprise and white collar. So basically, they are promising to enable private deployment so that enterprises and governments can keep their customers' data safe and behind their own firewalls, right? So this isn't a deployment where you're accessing their API. This is going to be on your own servers, your own firewalls protected. And that's kind of what coheres. Basically, I think it's the angle they're taking so that they can stay competitive in the market. This is what they said. This is a co-founder, Nick Frost of Frost. That's the greatest last name of a Canadian. I love that. He's a co-founder of Cohere. He said, LLMs are only as good as the data they have access to. If we want LLMs to be as useful as possible, they have to have access to useful data. And that means they need to be deployed in the customer's environment. ([Time 0:02:49](https://share.snipd.com/snip/8f5c5e37-092a-400c-ad50-c5b91f665b9e)) - **Keep Models On Your Infrastructure** - Deploy agents on your own infrastructure if you must protect sensitive data. - Run North in on-prem, hybrid, VPC, or air-gapped setups to keep data inside your firewall. Transcript: Jaeden Schafer And then when you ask it questions, it's actually looking at your data to give you results and responses. It's very customized and it's stuff that you can't obviously get publicly on something like ChatGPT. So instead of using enterprise cloud platforms like Azure or AWS, you can actually install North on your organization's private infrastructure so that it never sees or interacts With your customer's data. North can run on an organization's on-premise infrastructure, their hybrid cloud, their VPCs, or air-gapped environments. So this is pretty cool. He said, quote, we can deploy literally on a GPU in a closet that they might have somewhere. ([Time 0:04:12](https://share.snipd.com/snip/93870ac3-9cc3-4d50-be7b-7104f06af11d)) - **Security Controls Target Regulated Customers** - North bundles enterprise security features like granular access control and continuous red teaming. - These controls aim to meet compliance standards such as GDPR, SOC 2, and ISO 27001 for regulated customers. Transcript: Jaeden Schafer Basically, they claim that they also include security protocols like granular access control, agent autonomy policies, continuous red teaming, and third-party security tests. So basically, it meets international compliance, GDPR, SOC 2, ISO 27001. I'm not sure what country that is required by. And I'm sure some people are yelling at me that it's like a super common thing I just wasn't aware of. But in any case, this is pretty exciting. ([Time 0:04:50](https://share.snipd.com/snip/a34b85a1-fb1e-44a5-93ce-27dc90dda2d8)) - **Big Pilots Signal Enterprise Interest** - Cohere piloted North with big customers including RBC, Dell, LG, Ensemble Health, and Palantir. - These early pilots signal enterprise interest in private, compliant agent deployments. Transcript: Jaeden Schafer They said that they already piloted North with a bunch of customers. So RBC, Dell, LG, Ensemble Health Partners, and also Palantir. So there's some big players that are using this. It is a cool tool. ([Time 0:05:30](https://share.snipd.com/snip/f72b3333-10c3-4d57-a024-0b5aa60c6ec5)) - **Auditable Outputs Aid Trust** - North focuses on chat, search, and auditable reasoning chains for enterprise workflows. - Showing steps and citations helps employees verify outputs from internal data sources. Transcript: Jaeden Schafer It basically mirrors many AI agent platforms right out of the box. Its main feature is chat and search, which basically lets you get answers to your customer support queries. You can summarize meeting transcripts, you can write marketing copy, you can access information from internal resources and the web. So these are things that we have seen other places, but they're also basically trying to make this more focused on the enterprise and what they're doing in-house and not giving it like Access to, you know, they're not running this on their own server. You're running this on your personal company server. It also has citations and quote unquote reasoning chain of thought. So an employee could audit or verify the output that is actually there, right? It will show the steps that it went through to get an answer. ([Time 0:05:43](https://share.snipd.com/snip/ce4eb863-d936-4409-b3ce-fbffee3646b4)) - **Agents Create Assets And Research** - North combines generation and multimodal search (Compass) to create assets and research. - Features include tables, documents, slides, and higher-level market research automation. Transcript: Jaeden Schafer The chat and search function are basically powered by Cohere. They have their own LLN model. We should go try on AI Box if you're curious on how that works and what it looks like. Command, they have a bunch of generative AI models. Compass is its multimodal search tech stack that they've created. It's so funny. I love this. Compass North. This is great. This is very, very Canadian. So this is what they said about it. They said, it goes beyond just Q&A and gets into doing work for you. North has a bunch of asset creation. It can make tables. It can make documents. It can make slideshows. A bunch of market research, or it can do a bunch of market research. ([Time 0:06:41](https://share.snipd.com/snip/00c28c52-eda9-4e77-9b73-9497a502f1ca)) - **Autogrid Acquisition Boosted Features** - Cohere acquired Autogrid, a Vancouver firm that built enterprise market-research automation. - That acquisition helped accelerate North's market-research and automation features. Transcript: Jaeden Schafer I think it is worth noting that Cohere actually acquired a company back in May called Autogrid. So this was a Vancouver-based company. And basically, they develop enterprise tools for automating high-level market research. ([Time 0:07:17](https://share.snipd.com/snip/f7bc8f37-c314-4d18-a148-a0cfb1e520c8)) - **Connect Agents To Existing Tools** - Integrate agents with your existing workplace apps to unlock practical workflows. - Connect North to Gmail, Slack, Salesforce, Outlook, and MCP servers for enterprise automation. Transcript: Jaeden Schafer Like a lot of AI agent platforms, it can connect to existing workplace tools. So things like Gmail, Slack, Salesforce, Outlook, all of those, it can integrate with all of them. It also integrates with MCP, Model Context Protocol servers to basically access a bunch of industry specific or in-house applications that a company might have. ([Time 0:07:44](https://share.snipd.com/snip/d7a294a0-47b0-4646-8767-e604f8b58a29)) - **Tools, Not Just Model, Drive Value** - Cohere positions itself as a practical enterprise-stack player rather than the outright top LLM provider. - Its value relies heavily on tools and integrations around a competent model rather than model leadership alone. Transcript: Jaeden Schafer Frost himself from Canada. He said, as you build confidence by chatting to the model, there is like a smooth transition that happens between using this as an augmentation to using it as an automation. All right. Will you completely automate everything in your life with this platform? I will say probably not for now. I think this is probably a really useful tool, especially for companies. I don't think it's competitive as a standalone product. This is my critical after giving them their praises for their clever names and being Canadian. But I don't think this is actually a super competitive product as a standalone product. And I think this is evidenced by the fact that they're not competing head on with OpenAI and a lot of other players trying to have like the best model. They're basically trying to say, look, we have a really decent model and we have a whole bunch of really great tools for enterprise to use. Is this a bad approach? Not at all. I think this is a great approach. They're solving a problem that people have. I just don't think it's like, it's not something that everyone has to use or try because it's probably not number one. Now, is it good at certain tasks, better at maybe tone or better at style or better at doing different things? Yeah, I think it is. But I also think it's probably getting most of its lift and usage. It's most of its valuation based on the software tools that have been basically built around their LLM. And I think it almost calls into question, unless they're about to train and release like some new insane LLM, they're not number one. ([Time 0:08:07](https://share.snipd.com/snip/c2aa1430-9d9d-4b44-96f7-a6b26ee3463b))