The Futurist

MiniMax's cloud sandbox AI that builds skills from every task

“The thesis MaxHermes is betting on: within 2-3 years, enterprise AI value shifts from model capability to accumulated task memory — the agent that has already learned your workflows is worth more than the smarter agent starting fresh. That's a falsifiable, specific bet, and the self-evolving skill library is the technical mechanism for it. The second-order effect, if this works, is that switching costs in enterprise AI compound over time exactly like CRM data lock-in did in the 2000s — the longer you run MaxHermes, the harder it becomes to migrate because your skill library is proprietary. The trend line is the shift from stateless LLM calls to stateful agent infrastructure, and MaxHermes is early on it — the China-first integration set is a constraint today but a strategic beachhead if MiniMax's enterprise market share in APAC grows. The dependency that has to hold: skill extraction has to produce genuinely reusable abstractions, not just logged task histories, which is a hard ML problem they haven't proven publicly.”

Alibaba's open-source personal assistant that runs on your machine across every chat app

“Personal AI assistants that you fully own, run locally, and connect to every communication channel you already use — this is where the market is heading. QwenPaw is one of the most complete implementations of this vision available as open source today.”

A personal AI with persistent memory that plans and acts for you

“AI-to-AI social coordination is the sleeper feature here — the idea that your agent and a friend's agent can negotiate and plan together without either of you micromanaging is a genuinely new interaction paradigm. This is the early prototype of something that will be normal in 3 years.”

Open-source AI chat with enterprise RAG that runs anywhere

“Onyx represents a critical counter-movement to AI centralization. As enterprise AI spending scrutiny intensifies, self-hostable alternatives with full data sovereignty will capture the compliance-sensitive markets that hyperscalers are locked out of.”

Anthropic's AI assistant — best-in-class coding, reasoning, and computer use

“Extended thinking is a different cognitive mode — watching Claude reason through hard problems in real-time lets you course-correct before it goes wrong. Anthropic's safety-first approach is becoming a competitive advantage as trust in AI systems matters more.”

OpenAI's flagship AI assistant — multimodal, reasoning, and now video

“The memory feature compounds — the longer you use it, the more personalized it becomes. Projects make ChatGPT a persistent collaborator, not a stateless chat window. OpenAI is building the ambient AI layer and ChatGPT is the front door.”

Confidence-weighted AI ensemble that topped Humanity's Last Exam

“Confidence-weighted ensembling is the quiet breakthrough everyone is sleeping on. Individual models plateau — but smart aggregation keeps pushing the frontier. Sup AI scoring 52% on Humanity's Last Exam when no single model breaks 40% proves the thesis.”

An operating system that is pure AI

“This is the most ambitious rethink of computing I have seen since the iPhone. Ditching the file-and-folder paradigm entirely for AI-first interaction is either visionary or insane — probably both. If even 20% of this vision works, it will influence every OS built after it.”

Let 200+ AI models debate your question

“Multi-model deliberation is how we will make important decisions in five years. Seeing where models agree gives you real signal — and where they diverge reveals your blind spots. AI Roundtable makes this accessible to anyone right now.”

Inflection's personal AI — empathetic and conversational

“Pi represents a different AI future — not about productivity but about human connection. As AI companions become normalized, Pi has first-mover advantage in emotional intelligence.”

xAI's unfiltered AI with real-time X data

“Having real-time social data baked into an AI is unique. For trend analysis, market sentiment, and cultural pulse-checking, Grok fills a niche no one else does.”

Google's multimodal AI with Deep Think reasoning

“Google's advantage is integration — Gemini in Gmail, Docs, Meet, Maps. When AI is everywhere in your workflow, the compound value is enormous.”

AI agent orchestration platform

“Production AI agents require infrastructure that handles failures gracefully. Inngest is building exactly that.”

Model Context Protocol for AI tool integration

“MCP is becoming the standard for AI-tool integration. The protocol approach scales better than point-to-point integrations.”

Standard library of AI tools and integrations

“A standard library of AI agent tools will become as essential as standard libraries for programming languages.”

Integration platform for AI agents

“The integration layer for AI agents is essential infrastructure. Composio's breadth of integrations creates a real moat.”

Self-hosted AI interface

“Self-hosted AI interfaces will be standard for privacy-conscious users and organizations. Open WebUI leads here.”

Memory layer for AI applications

“Persistent AI memory is a missing piece for meaningful AI assistants. Mem0 is the leading solution in this space.”

Framework for orchestrating AI agents

“Multi-agent orchestration will be essential as AI tasks grow more complex. CrewAI's simplicity gives it adoption advantage.”

Open-source ChatGPT alternative that runs offline

“Desktop AI apps that run locally will be a major category. Jan is building the consumer interface for local AI.”

Microsoft's multi-agent conversation framework

“Microsoft Research backing and enterprise integration path make it the safe bet for enterprise multi-agent systems.”

Open and efficient AI models from Europe

“European AI sovereignty matters. Mistral proves world-class AI doesn't require US hyperscaler resources.”

Unified API proxy for 100+ LLMs

“Multi-model architectures need a proxy layer. LiteLLM is becoming the standard infrastructure for LLM routing.”

Programming — not prompting — LMs

“The idea that prompts should be compiled, not handwritten, is correct. DSPy is ahead of its time.”

AI gateway for production LLM apps

“AI gateways will be standard infrastructure. Portkey's focus on reliability and guardrails addresses real production needs.”

Unified API for every AI model

“Model diversity will only increase. A unified API layer becomes more valuable as the model landscape fragments.”

State-of-the-art embedding models

“Domain-specific embeddings will become standard. General embedding models leave performance on the table.”

Microsoft's AI orchestration SDK

“Enterprise AI adoption will go through existing stacks. Semantic Kernel meets .NET developers where they are.”

Data framework for LLM applications

“Data integration is the real bottleneck for enterprise AI. LlamaIndex is correctly positioned at this chokepoint.”

Framework for developing LLM-powered applications

“Despite the criticism, LangChain's ecosystem (LangSmith, LangGraph, templates) is the most complete platform for LLM apps.”

Create and chat with AI characters

“Character.ai has the best understanding of long-context character consistency. That tech could be transformative if applied elsewhere.”

Computer vision infrastructure

“Computer vision is expanding beyond traditional use cases into real-time analysis. Roboflow's platform scales with this growth.”

Enterprise AI with RAG specialization

“Cohere's enterprise focus and RAG specialization create a defensible niche in a market dominated by generalists.”

Data labeling and curation platform

“Data quality is the bottleneck for AI. Labelbox addresses the most important constraint in model development.”

ML experiment tracking and model registry

“As AI development becomes more systematic, experiment tracking becomes foundational infrastructure. W&B leads here.”

Data engine for AI

“The data engine for AI is as important as the compute engine. Scale's position in frontier model training is unique.”

The AI community building the future

“The open-source AI hub will only become more important as the model ecosystem grows. Hugging Face has the network effects.”