Select Page

The 10 Camps of AI Innovation

by | Feb 27, 2025 | AI, Video | 0 comments

AI Innovation for Beginner’s (Like Me)

As I discussed in my last post, my LinkedIn feed is a firehose of AI hype—much of it noise, some of it genuinely game-changing. The pace is overwhelming, but one thing is clear: we either adapt to AI or risk irrelevance in digital work.

So where should we start? We can’t research every shiny tool since there are dozens released every day. Instead, I’m zooming out and mapping AI into 10 categories or “camps” to manage the chaos and see the bigger picture. And I’m not an AI engineer or expert – if you are looking for deep diver AI tips, I’d follow Chip Nuyen and Nathan Lambert. 

Also if you are more interested in using AI tools ASAP, my advice is to start with a clear objective—what do you want to accomplish? For example: I want to build an app that tracks tech industry M&A. Use YouTube to do research and then test what you find. I was amazed at how easy it was to test AI-powered app builders like Bolt.new and Bubble.ai, and also tested automation tools like Relevance AI and Zapier for creating AI agents. It’s not rocket science—just define your goals and experiment. 

In my next post, I’ll dig into how AI agents can automate product marketing—because if AI can tackle rocket science, it can certainly handle PMM tasks. Stay tuned for that. 

Methodology

My approach to this research was to first take notes on every post or innovation that I came across where a tool / framework / platform was mentioned. This mostly came from social media, especially LinkedIn, but also (naturally) by asking ChatGPT for help. Then I began taking a huge list and visited every website to learn more. I created accounts then I broke things, had fun and got silly. I didn’t spend a lot of time on each tool, just enough to say, “Oh, I see now.” Before I got lost in the vortex, I broke down my findings into key categories—or “camps”—to help make sense of the chaos. 

Here’s the result. A list that highlights the 10 major AI camps and their roles in shaping the AI landscape.

Pro tip: For every vendor you aren’t familiar with below, simply search for their brand on LinkedIn and click Follow. Then just observe their releases and community engagement. I have been amazed at what they are up to. 

The Ten Camps of AI Innovation

1. Chip Makers: The Foundation

I know this one is a no-brainer, but chip design and manufacturing forms the backbone of AI innovation. Companies like NVIDIA, AMD, and TSMC produce GPUs and processors that power AI workloads, needed for training large language models (LLMs) and running complex AI systems. 

NVIDIA, in particular, leads this space with its cutting-edge AI-optimized chips, making it the tip of the spear for AI hardware and software development. If you haven’t watched Jensen Huang’s CES or GTC keynotes yet, you need to go check them out so see how they are at the forefront.

2. Large Language Models (LLMs): The Brains Behind AI

A Large Language Model (LLM) is a type of artificial intelligence (AI) model designed to understand, generate, and manipulate human-like text. LLMs are built using deep learning techniques, particularly transformer architectures, and are trained on massive datasets of text from diverse sources, including books, articles, and online content. LLMs power chatbots, AI assistants, content creation tools, and research applications, playing a crucial role in modern AI-driven automation and decision-making.

Some of the most prominent LLMs include:

  • OpenAI – ChatGPT: Scheduled for launch within weeks, GPT-4.5 will be the final model without full chain-of-thought reasoning, a method that enhances complex problem-solving by generating intermediate steps. Following this, GPT-5 is planned for release in the coming months, aiming to integrate various OpenAI technologies, including the o3 model, to create a unified AI system.
  • Google – Gemini: Google DeepMind has unveiled “Gemini 2.0,” an advanced AI model designed for the agentic era. Gemini 2.0 offers multimodal output capabilities, including native image generation and audio output, and integrates seamlessly with tools like Google Search and Maps.
  • Anthropic – Claude: Anthropic has released the “Claude 3” model family, comprising Claude 3 Haiku, Claude 3 Sonnet, and Claude 3 Opus. These models set new industry benchmarks across various cognitive tasks, with Claude 3 Opus being the most advanced, excelling in complex analysis, higher-order math, and coding tasks.
  • Meta – Llama: Meta has introduced “Llama 3.2,” the latest in their open-source AI model series. Llama 3.2 includes small and medium-sized vision LLMs, as well as lightweight, text-only models optimized for deployment on edge devices, enhancing accessibility for developers.

I highly recommend the videos by Andrej Karpathy to learn more. He is the top resource on simplifying LLMs for beginners. 

These LLMs dominate the conversation because of their early adoption, daily usage, and ability to rapidly release groundbreaking features. A single feature launch—like ChatGPT Operator—can disrupt entire software ecosystems by rendering thousands of apps obsolete.

3. AI-Native Startups: The Application Layer

Using the above LLMs, AI-native startups are spearheading innovation by building new apps and tools on AI-driven foundations. Many leverage AI coding assistants which drastically lower the barrier to entry for app development. Apps no longer cost hundreds of thousands of dollars or hiring engineers to build. 

I personally signed up for each of these tools. I was using Bolt.new for maybe 2 minutes before I had the interface coded for an app idea I’ve had for 15+ years. Now, I (and you) have the power to build apps in minutes, without any coding knowledge. 

It’s really amazing, but one catch was that once I decided to move forward and keep building this app, Bolt shrewdly ended my supply of free tokens and tried to upsell me. Can’t blame them, they have VC’s to repay. I declined, as I figured there were other tools to explore, which there were. Moving to Bubble.io, I built an app as well, and I’m testing that some more. Tools like these are giving everyone the ability to build apps in minutes—no coding required. But while this flood of innovation is exciting, it’s also flooding the market with new apps by the hour.

4. Existing SaaS Providers Adding AI Assistants

Established SaaS providers, such as Microsoft adding Copilot, HubSpot AI and Salesforce Einstein, are baked-in AI assistants inside of their already-adopted applications as a survival strategy. This is just to name a few but I think you would be hard-pressed to find a SaaS vendor that isn’t doing this. 

Why do I say survival strategy? Am I being overly dramatic? My view is that these vendors are facing a major threat from AI-native startups. For example look at the emergence of copy.ai, a once-tiny startup that is now on Salesforce’s radar in this area called, “AI GTM”. Just look at both of their new positioning statements, and they look very similar. Of course there can be no comparison when it comes to scale, users and capacity – but look at the reality these days, many companies are operating with much less people thanks to AI and automation, so we shouldn’t discount a company just because they are “small.” 

There are many “Salesforce’s” facing widespread disruption from the masses of AI startups entering the scene, and are scrambling to innovate and keep talent in house. Another example is simplified.com providing AI bots across the board for marketing teams. 

I picture the current AI startup landscape as a game of Pac-Man—but with a twist. Right now, AI startups are like Pac-Man, racing through the market maze, trying to gobble up market share from established SaaS vendors. But in reality, the game may play out differently. Instead of startups overtaking the incumbents, we’re more likely to see established players in the Pac-Man role, acquiring companies like Copy.ai and Simplified.com rather than building their own solutions from scratch. In the end, it may be a game of buy vs. build, with the bigger fish swallowing the smaller ones.

Another major threat to existing SaaS vendors is LLMs integrating core SaaS functionalities. If AI models like ChatGPT can introduce features like Operator, enterprise-grade capabilities won’t be far behind. Imagine an Operator for SaaS—instead of juggling 10 different SaaS subscriptions, users could rely on AI assistants that dynamically call APIs, retrieve data, automate workflows, and replace entire B2B software categories. The shift from static SaaS platforms to AI-driven automation could fundamentally reshape the enterprise software landscape.

5. Enterprise AI Agent Development Frameworks 

This is where the practical application of AI gets exciting. Developer-friendly tools and no/low-code platforms are making it easier than ever to build AI agents that interact with their environments. Many of these are focused on enterprise use cases because that’s where the big money is, and also happens to be where much of the demand is. Examples include:

  • LangChain: Framework for connecting LLMs with external data sources and workflows.
  • LLamaIndex: A framework for ingesting, indexing, and querying enterprise data.
  • Rasa: Context-aware chatbot and assistant framework.
  • AutoGen (Microsoft): Open-source framework for multi-agent systems designed to handle complex workflows.
  • Semantic Kernel: Framework for structured AI workflows.
  • Haystack: Open-source NLP framework for question-answering systems.
  • Letta AI: Lightweight open source framework for building stateful agents that “live forever”. 

These frameworks also enable enterprises to build in-house AI systems to process private datasets—critical for companies with privacy concerns. Think of a company with 10 million internal documents wanting to train AI agents to query and analyze that repository in a secure way. These frameworks make it possible.

6. User Friendly AI Agent Development Platforms

There are two sub-camps for AI Agent Development Platforms, the advanced level platforms which I just went over, and the simpler, user-friendly platforms that are built for less advanced users that want a quick start. I found myself getting lost when trying out LangChain and the others above, because they are more geared toward internal enterprise engineering teams in my view. But the following tools are user-friendly platforms that beginners can use to quickly build AI agents. These are much easier to use and more geared for individual app developers. 

I tried Relevance AI and built an agent with it very quickly. Within that agent, I could define Subagents and plug into an ecosystem of tools. I will share more on this in my next post when I explore the creation of AI agents for product marketing.

7. Enterprise-focused AI Agent Development Platforms

While most tech giants are deploying their own LLMs (e.g. Google Gemini and Microsoft’s partnership with OpenAI), some are simultaneously creating dedicated agent development environments to help developers within enterprises build their own AI apps and agents. For example I found the following platforms:

  • Google Vertex AI Agent Builder: A no-code tool for developing agents with natural language capabilities.
  • Microsoft Azure OpenAI Service and Copilot for Developers (GitHub Copilot)
  • IBM watsonx.ai Agent Builder: Simplifies AI agent creation with prebuilt templates.
  • Amazon Bedrock: service from AWS that allows developers to build AI applications using LLM models. 
  • Cohere: Embedding APIs and natural language tools for text classification and generation.
  • Pinecone: Stores memory for AI agents, enabling contextual understanding.

These platforms are significant players for enterprises looking to integrate AI agents into their existing ecosystems.

8. Open Source Machine Learning Hubs and Frameworks

Open-source tools provide developers with flexibility to build, customize, and deploy machine learning models. Examples include:

  • Hugging Face: Open-source platform and community hub for sharing, collaborating and developing machine learning models and tools.
  • TensorFlow: Open-source framework for machine learning to create ML models.
  • PyTorch (Meta): By Meta, an open source deep learning framework. 

These tools give developers control and customization options, making them ideal for enterprises with specific AI needs.

One of the more interesting sites that I discovered was a platform and community hub for machine learning called huggingface.co. I learned there are 1.43 million+ LLMs/models in development, which seems to be growing by about 100,000 models every few weeks. They also filter their models by the following categories:

  • Tasks: This filters models by the “mode” of input. So audio-text-to-text or video-text-to-text, etc. 
  • Libraries: This filters by frameworks and the underlying technology the models are built with such as PyTorch (Meta) and TensorFlow. . 
  • Datasets: This filters by machine learning datasets provided by Mozilla and many others. 
  • Languages: This allows you to filter models by language requirements. 
  • Licenses: This filters by licenses which define how software can be used, modified or shared. 

9. AI Automation Platforms

Automation platforms are the glue of the AI ecosystem, bridging workflows across tools and systems. By connecting everyday apps with AI agents, these platforms make automation accessible:

  • Zapier: Connects apps and triggers actions automatically.
  • Make (formerly Integromat): Visual tool for designing and automating complex workflows.
  • Heyboss.xyz: Customizable AI agent solutions for business workflows.

For example, you can connect Google Sheets row additions to LinkedIn drafts to automate social media posts. If you’ve noticed an uptick in LinkedIn activity recently with people posting perfectly written content on new topics, these tools are why! 

10. AI Simulation & Autonomous Agents

AI isn’t just about understanding language—it’s increasingly about taking action in digital and real-world environments. The future of AI innovation lies in autonomous agents that can reason, plan, and execute tasks with minimal human oversight.

Key areas driving this shift:

  • Decision-Making AI: Agents that use reinforcement learning (e.g., OpenAI Gym and DeepMind’s AlphaZero) to make strategic choices.
  • AI in Simulations: Tools like NVIDIA Omniverse and Unity ML-Agents allow AI to learn by interacting with virtual environments.
  • Multi-Agent Systems: AI agents that coordinate, negotiate, or compete, leading to emergent intelligence in fields like gaming, robotics, and finance.
  • Autonomous Task Execution: Projects like AutoGPT, BabyAGI, and Microsoft’s Jarvis show the potential for AI to work across multiple applications without direct prompts.

As AI moves from being a passive tool to an active participant in robotics and workflows, these innovations will define the next generation of self-learning, self-optimizing systems that don’t just answer questions—but solve problems on their own.

Conclusion

So what’s next? I’ll keep exploring, testing, and breaking things to make sense of this chaos—and I’d love to hear your take. What tools are you using? What’s overwhelmed or inspired you lately? Let’s figure this out together.

Please note that the next post on this topic will look specifically at the use of AI apps and agents to optimize product marketing workflows. Will I find anything real or will all of this just be more noise without practical applications? Let’s see, I’m eager to find out. 

Useful Research Links

⤵️ Join the conversation with 300+ other Cyber PMMs