Map of the AI agent industry
Update: Oct 23, 2026. Authored: Oct 19, 2026.
Orientation
AI tools are complex and the tools required to make an AI product are diverse. Agents like Bronco sit upon all the other foundational technology and directly interface with users.
Agents are traditional computer programs that orchestrate lots and lots of AI inferencing work to solve human problems. They filter the output, make it safer and more predictable, and help you control the power of AI.
Inferencing is the actual calculation of AI. Its how a computer converts electrical energy from your wall outlet into a useful answer to a problem.
Weights are the data that's created when an AI is trained. They store the useful information an AI has. Training models to produce weights involves lots of energy and time.
And lastly hardware provides the literal foundation of AI. These are specialized computers that combine user input with weights during inferencing. The hardware is controlled by the agent.
Agents
Agent developers build the tools that let models read files, run commands, edit code, and verify their work.
Inferencing
For agents, inferencing is the process of combining user data with LLM data to generate new content and operate computers.
Weights
Model developers train base models and tune them into assistants, and publish open weights that anyone can download, run, fine-tune, and ship.
Hardware
Inferencing is a computer calculation and hardware providers build the computers that do it.
Sources
Each source lists the date it was retrieved. The data notes behind the illustration are longer than this post; this list covers the claims in the text.
- OpenRouter, LLM Rankings. Week of October 12, 2026 text-request shares, including DeepSeek at 24.4%. Retrieved October 19, 2026.
- OpenRouter, Embedding Model Rankings. Week of October 5, 2026 embedding and rerank shares, including OpenAI at 49.7% of embedding requests and Voyage at 75.3% of rerank requests. Retrieved October 19, 2026.
- Vercel, AI Gateway Production Index, October 2026. Open-weight models at 71.5% of tokens and 23% of spend; DeepSeek at 58% of tokens and 6.6% of spend. Retrieved October 19, 2026.
- Cubic, Best AI coding tools and agents, with Bug Bench data. Week of October 5, 2026 sample: Claude Code 78.1% and Codex 18.8% of AI-written code reviewed. This is a vendor sample, not an industry-wide measure. Retrieved October 19, 2026.
- GitHub repository pages, retrieved October 19, 2026: Ollama, vLLM, SGLang, MLX, OpenCode, Gemini CLI, Cline, Aider, Continue.
- Qwen, Qwen3-Embedding-0.6B model card. Sizes, context, dimensions, license, and the 8B model's 70.58 MTEB multilingual score as of June 5, 2025. Retrieved October 19, 2026.
- Model Context Protocol, official site and specification repository. The 2026-07-28 revision and the stateless design. Retrieved October 19, 2026.
- A2A Protocol, official site, and Agno, A2A vs MCP. A2A v1.0 in March 2026, 150+ supporting organizations by April 2026, and Linux Foundation governance. Retrieved October 19, 2026.
- AGENTS.md, official site, and the Agentic AI Foundation. The format, the 60,000+ examples, and AAIF stewardship of MCP, A2A, and AGENTS.md. Retrieved October 19, 2026.
- Synergy Research Group, Q2 2026 cloud infrastructure estimates, as reported by Tech Insider. AWS 28%, Azure 20%, Google Cloud 15%, $143.4 billion quarterly market, 43% year-over-year growth. Retrieved October 19, 2026.
- Nvidia, financial results for the fourth quarter of fiscal 2026. $62.3 billion in data center revenue, up 75% year over year. Retrieved October 19, 2026.
- Nvidia, OpenAI and NVIDIA announce strategic partnership to deploy 10 gigawatts of NVIDIA systems, September 22, 2025. Retrieved October 19, 2026.
- Safeguard, OpenAI's compute plan, citing AMD and OpenAI's October 6, 2025 announcement. The 6-gigawatt AMD commitment and the first 1 gigawatt of MI450 in the second half of 2026. Retrieved October 19, 2026.
- Safeguard, AI chip market share 2026, citing Silicon Analysts, February 2026. Nvidia at 75% of AI accelerator revenue in 2026, down from 81% in 2025. Retrieved October 19, 2026.
- THE D*AI*LY BRIEF, Together vs Fireworks vs Groq. DeepInfra's October 17, 2026 Kimi K3 pricing of $2.85 input and $14.25 output per million tokens, against $3 and $15 at Moonshot, Together, Fireworks, Baseten, and Bedrock. Retrieved October 19, 2026.
- LMSA, Open-Weight LLM Leaderboard, last updated October 10, 2026. Retrieved October 19, 2026.
- MTEB leaderboard, LMArena, SWE-bench, and Artificial Analysis. Retrieved October 19, 2026.
- Mixpeek, Best Vector Databases in 2026. A comparison of managed, open-source, and database-native vector search, including Pinecone, Qdrant, pgvector, Milvus / Zilliz, Weaviate, turbopuffer, Amazon S3 Vectors, LanceDB, Chroma, Vespa, and Elasticsearch. Retrieved October 19, 2026.
- Apple, MacBook Pro tech specs. Up to 128GB of unified memory with M5 Max. Retrieved October 19, 2026.
- Apple, Mac Studio. Up to 512GB of unified memory with M5 Ultra. Retrieved October 19, 2026.
- Nvidia, DGX Spark. GB10 Grace Blackwell superchip with 128GB of unified memory. Retrieved October 19, 2026.
- Nvidia, RTX Spark. Up to 128GB of unified memory and 1 petaflop of FP4 compute. Retrieved October 19, 2026.
- Layer3Labs, Best AMD Ryzen AI Max+ 395 128GB Mini PCs (2026). 128GB of unified memory with up to 96GB assignable to the integrated GPU. Retrieved October 19, 2026.
- Liliputing, AMD launches Ryzen AI Halo mini PC with Ryzen AI Max+ 495 and 192GB RAM. Retrieved October 19, 2026.
- Tech Insider, Surface Laptop Ultra $2,599 AI PC: Local LLM Specs, and Surface RTX Spark Dev Box: $5,999, 128GB RAM. Retrieved October 19, 2026.
- OEM coverage: Tom's Hardware on the Sipeed MaixStation, StorageReview's best laptops for local AI, and Framework Desktop. The OEMs and 128GB-class machines named in the text come from these launches and reviews. Retrieved October 19, 2026.
- Fixture, Agentic CRM and GTM platform. Product description and agent capabilities, retrieved October 22, 2026.
- Notion, Agents, rebuilt to do more for less. Notion Agents with their own computer and browser, tasks up to 24 hours, and the Auto model router, retrieved October 23, 2026.
Changelog
- October 22, 2026: Added "Fixture" Agentic CRM.
- October 23, 2026: Added "Notion" Agent.