Remix's GenAI Tech Stack
A rundown of what's in our toolbox
After a year and a half of helping SMBs implement GenAI, we’ve learned that falling in love with a single tool is like trying to build a house with just a hammer. Sure, you can pound everything into submission, but you’ll miss the precision of a saw, the finesse of a chisel, and the efficiency of a drill.
This week, we’re opening our tool box and showing you exactly what we use at Remix Partners, what we love (and hate), and how they work together.
📰 What’s Happening in GenAI
Everything Works with Everything
The three primary frontier model families now connect to the two major email / shared drive systems. ChatGPT, Claude and Gemini can now connect to Microsoft 365 as well as Google Workspace. ChatGPT, Claude, and Gemini can all access your files whether they’re in SharePoint, OneDrive, or Google Drive. This means your choice of AI no longer depends on your office suite; you’re free to pick based on capability, not compatibility.
Connect Gemini to M365 >
Connect Claude to M365 >
Google Released Gemini 3
Google launched Gemini 3 on November 18th. Based both on its scores on a slew of benchmarks, as well as the vibes, this model seems quite impressive! It is still too early for Remix to have a strong POV, but we’re planning to test this model on a couple of projects over the next few weeks.
Read more >
Google Flows: Promising but Proceed with Caution
Google Workspace now has a (limited) agent builder baked in called “Flows”. Create an email draft when a document is shared, auto-summarize meeting notes to Chat, or generate weekly reports from your spreadsheet data. Flows can connect to pretty much every Google Workspace service, and a few external tools as well, like Asana,Salesforce, and Quickbooks. Unfortunately, they’re failing about 25% of the time in our testing with no clear error messages. Experiment, but don’t use them for anything mission-critical yet. Note that all of our testing was before the launch of Gemini 3, which 🤞improves things.
Read more >
Agent Skills Give Claude Superpowers
Claude has launched “Skills,” which is - in a way - their version of a Custom GPT. Dig into your settings and enable them, especially the Skill that creates Skills! We’ve transformed a number of our projects into skills, which empowers Claude to leverage them no matter which project you’re in, and only when that skill is needed. For example, we’ve turned our Claude “Proposal Writer” project into a skill, which makes keeping cleanly organized client data massively easier in Claude.
Read more >
Remix’s GenAI Stack
Frontier Model Promiscuous
Model monogamy might be holding your business back.
We maintain subscriptions to Claude, ChatGPT, Gemini, Copilot, and Grok. Per person, this costs us about ~$250/month (note: this includes the $100/person/month Claude Max plan). We know our time is valuable, and the return on investment is incredibly obvious for us.
Each model has distinct strengths:
Claude: Writes like a human, reasons through complex problems, longer context window
ChatGPT: Excellent researcher, top-notch data analyst, shortest context window
Gemini: Deep integration with Google Workspace, massive context window
Grok: Real-time information from X
Despite repeated attempts, we’ve been unable to find a practical business application of Meta AI. Maybe this will change in the future; it would be great to have yet another tool in the toolbox. Microsoft Copilot provides access to both ChatGPT and Claude, but is not otherwise a frontier model.
The learning curve is real. But once you develop model fluency, you can match the right tool to each task and save hours each week. On any given day, we’re likely to use at least three of the frontier models, and there are frequently days we’re in all four.
Below are all of the primary GenAI tools that we use, what we love about them, what drives us crazy, and a quick example of how they show up in our work. If you’re inspired to try one of these platforms, note that some of these links are to referral programs that give you a discount, and a small fee to Remix..
Claude
What we love: Projects keep context organized across client projects. Claude has the second-biggest context window among the models we use and is great at working on longer documents (especially with Opus 4.1). Skills are incredibly easy to create and Claude knows when to use them without loading the whole skill, preserving space in the context window. I find its integration with Google Workspace better than Gemini (caveat: I’ve not re-tested this with Gemini 3). Its writing quality is unmatched; it sounds like a thoughtful colleague, not a robot. MCP on the desktop app opens up a world of possibiltiies.
I’m on the Max plan ($100/month) which means no more hitting rate limit walls at 2 PM and having to stop work. Worth every penny if you’re using Claude for a few hours per day.
What drives us crazy: No scheduled actions, and it can’t generate images or videos (although it can understand images quite well). The rate limits on the base plans can be extremely frustrating.
An example of our Claude workflow: We use Asana for tasks, Attio for our CRM, and Google for calendars and emails. I’ve got a Claude project that’s my “Chief of Staff” which, when prompted to “give me the rundown,” will look across my calendar, emails, tasks, and pipeline to help me prioritize the most important actions I should take - adjusted to the amount of heads-down time I have that day.
ChatGPT
What we love: ChatGPT 5.1 is an incredibly capable model; most notably, it is a better, more human-sounding writer than ChatGPT 5. Its deep research capability, while not as great at uncovering sources as Gemini, seems to be better at analyzing research and finding unique insights. Several of our client projects from 6+ months ago live here, and ChatGPT works well enough that migration isn’t worth the effort; for much of 2024, this was our primary tool.
What drives us crazy: For 52 days, we attempted to get a human at OpenAI to resolve an issue when we migrated the domain on our account. Their bot-based support highlighted that it could not fix my issue and promised to tag in a human. For 52 days, no human replied, until literally yesterday. Our issue isn’t resolved, but at least we’re one step closer. It’s the single worst customer service experience I have ever had in a professional setting and reinforces the fact that the frontier labs just don’t care about small businesses.
Our ChatGPT workflow: We have a number of custom GPTs that are quite helpful, including a “write in my voice” one. I love to be working with all of the contextual documents in a Project, get to a document that is close, then “tag in” my custom GPT (by using the “@” symbol) and adjust the text to be in my voice without having to copy/paste my work across multiple projects and GPTs.
Gemini
What we love: Inside Google Docs, Sheets, and Slides, it’s unbeatable. Gemini’s notes and transcripts in Chat are top notch. The 1-million token context window means we never worry about running out of space. NotebookLM is our favorite research assistant; upload 20+ documents and it finds connections you’d never spot. Deep Research consistently surfaces better sources than other models. Gemini 3 is brand new, so we’ve not yet had a chance to really push it… stay tuned.
What drives us crazy: The writing quality (at least in Gemini 2.5 Pro) was noticeably more robotic than Claude or ChatGPT. It tended toward stale, corporate buzzword bingo (at least in our opinion). But we’re eager to try Gemini 3! Stay tuned. Gems feel rudimentary and have not seen an update in quite some time. Its ability to find things in Gmail and Calendar pales in comparison to Claude.
Our Gemini workflow: When analyzing a large set of documents, we dump everything into NotebookLM. Ask it to find patterns across all materials and it surfaces insights like, “your client mentioned budget concerns in 7 different contexts but never directly.” The automated FAQ generation has saved hours of work. This tool seems to get better on an almost weekly basis. .
Grok
What we love: Real-time access to the X/Twitter firehose. “I remember seeing a tweet about [some half-remembered link I clicked on], can you help me find it?” works shockingly well. It understands context and culture in a way other models don’t. Like it or not, the primary public GenAI conversation happens on X.
What drives us crazy: The different personalities that Grok has baked in are…interesting. I’ve described some as “a harassment lawsuit waiting to happen.” Be exceedingly judicious with how you use this model.
Our Grok workflow: It’s great at generating weekly trend reports on AI discourse. “What are people saying about [new development in GenAI land] and what are the main criticisms / points of discussion?” It synthesizes hundreds of posts into coherent themes, catching early signals we might otherwise miss.
Supporting Cast
Images: ChatGPT + Gemini Nano Banana
The images for this newsletter are made with ChatGPT’s image generation capabilities; it’s a workflow we’ve used for a while. But lately, we’ve been using Gemini’s “Nano Banana” image model (look for the 🍌 icon) and are quickly falling for it. The new version powered by Gemini 3 is best in class.
Transcriptions: Granola + Wispr Flow
Granola has quickly become one of my favorite GenAI tools. It doesn’t take up a tile in an online meeting, can record offline meetings, and the integration of LLMs into the interface is great. I regularly use it as a form of augmented memory; it stays open all day on my desktop. The only downside: backing up to Google Drive requires a separate Zapier automation for each folder (annoying).
Wispr Flow is a dictation app, although describing it that way hides just how useful it is. I am someone who talks much faster than I can type, so this empowers me to engage with technology at the speed of my brain. I especially love how it automatically switches from “spoken text” to “written text,” cleaning up stray words as well as adding proper punctuation and formatting.
CRM: Attio
I’ve held a number of business development roles and have been the decision maker who’s bought Salesforce, HubSpot, and Copper. Attio is the first CRM I don’t hate. Setup took less than an hour. The onboarding call was 30 minutes. It just works.
Vibe Coding: Replit
Every few months, I have an experience with GenAI that keeps me up at night thinking about the implications of what I just experienced. My most recent night like this came after a day of using Replit’s new Agent 3 “vibe coder”.
Turn on full autonomy, load up your credits (I’d start with ~$50), and it’s wild what you can build without writing a line of code. Its “secrets” feature ensures you don’t accidentally expose API keys, and the fact that you can design -> code -> security check -> deploy -> host all in one platform is freaking amazing.
Workflow Automation: Zapier + Google Flows
Zapier is fantastic in that it connects to pretty much everything and has an increasingly powerful automated workflow builder. Zapier is wildly frustrating in that you need to be somewhat technical in order to get a Zap to work perfectly, consistently. But we love it because it talks to just about everything.
Google Flows are the newest “agent” builder inside of Google Workspace. Our favorite one so far reviews all incoming emails, decides if that email is a receipt, forwards it to Quickbooks, and archives it. Super simple yet powerful. Unfortunately, we’ve had some of our more complex, multi-step Flows fail ~25% of the time (although this was pre-Gemini 3). These are certainly worth experimenting with, but still feels like prototype software.
We’re also playing around with the new AI features that have been added to enterprise apps like Notion, Asana, and Quickbooks. More on those in the future, but it’s so clear to us that the disruption of big enterprise apps by native AI apps and DIY vibe coding workflows is upon us. Sam Altman recently commented that we’re entering, “the fast fashion era of SaaS very soon.”
Networking: Happenstance.ai
While we won’t delete our LinkedIn account anytime soon, we increasingly first turn to Happenstance.ai to search our network. Imagine the powerful search tools that you find in a team LinkedIn Sales Navigator dashboard, but with a plain language chat-like starting point. It pulls data from LinkedIn, Gmail, your address book, and Twitter / X. We’re excited to see what they do next.
A Living Toolbox
In June, we wrote about how to think about picking a model family. Just six months later, things have radically changed. “Connectors” among the frontier models have exploded, and we’re rapidly approaching a world where everything can work with everything.
That’s equal parts exciting and overwhelming.
My grandfather was a carpenter, and I vividly remember going into his workshop and marveling at the array of hammers, screwdrivers, planes, and saws hanging on the wall. If asked, he would explain how each tool was there for a reason, with each bringing some speciality that the other tools didn’t provide.
Think about your own GenAI toolbox in this same way, with each tool that you bring into your work serving a specific purpose. Some will be more generalized, others highly specialized. Don’t hoard them, but know that a master craftsperson’s workshop usually has many highly specialized tools in it.
Over time, expect your tool set to evolve, driven by what you’re working on and the ever-evolving capabilities of the tools.
Questions? Email us at info@remixpartners.ai — we read every message.