YOUTUBE
The most valuable AI‑agent launches are those that become infrastructure for the tools your team already uses—offering seamless data access, open APIs, and stackability—rather than flashy, closed‑off demos. Apply the five‑question filter to decide whether a new agent worth an afternoon’s evaluation.
Nate B Jones argues that the AI‑agent market has shifted from “model‑centric” hype to “infrastructure‑centric” competition. The decisive factor is not benchmark scores but whether a launch integrates with existing workspaces, exposes critical data, supports ecosystems, and allows composability. He demonstrates this with five recent releases, showing which pass the filter and why.
Filter‑First Evaluation – A five‑question checklist (tool integration, openness, data ownership, ecosystem, stackability) quickly separates infrastructure‑level launches from feature demos. [^1][✓]
OpenAI Workspace Agents – Provide shared, cloud‑run agents that operate across ChatGPT, Slack, and scheduled workflows, ideal for recurring, cross‑tool team processes but less suited for native Salesforce or Microsoft 365 tasks. [^2]
Salesforce Headless 360 – Exposes every platform capability as an API, MCP tool, or CLI, turning Salesforce into agent‑ready infrastructure. With >60 new MCP tools and 30+ pre‑configured coding skills, it scores high on all filter dimensions, making it a must‑look‑at for RevOps teams. [^3][✓]
Microsoft Copilot Wave 3 (Co‑work + Work IQ) – Embeds agents directly inside Microsoft 365, granting deep data access and native permissions—excellent for purely Microsoft‑centric workflows, but relatively closed to external agents. [^4]
Moonshot Kimmy K 2.6 – Open‑weight, multimodal model with 300‑agent swarm; technically impressive but lacks the integration layer most enterprises need. Best for self‑hosted dev teams wanting full control over data and model, not for typical business users. [^5]
Perplexity Personal Computer (Mac) – Turns a local digital worker into a research‑centric agent with file‑system access and broad connectors. Fits research‑heavy, deliverable‑focused tasks (e.g., market analysis) but isn’t a shared workflow platform. [^6]
Layered Agent Strategy – Rather than “switching” from Claude, ChatGPT, or Copilot, organisations should layer agents: keep a default model for raw reasoning, use wrappers that provide the right data/graph, and add specialist tools where they win clearly. [^7]
“The question is not what launched this week. The question is now which of these millions of things actually deserves an afternoon of my team's attention.” — Nate B Jones [^1]
“Salesforce is not launching an agent. Salesforce is trying to become infrastructure under the agent economy.” — Nate B Jones [^3]
✓ VERIFIED — Salesforce Headless 360 includes >60 new MCP tools and 30+ pre‑configured coding skills. Confirmed by Salesforce’s official announcement (TDX 2026) and multiple industry reports. [^3]
⚠ UNVERIFIED — OpenAI’s Workspace agents can be scheduled. The public documentation mentions “scheduled runs” but lacks detailed API specs; verification pending.
⚠ UNVERIFIED — Kimmy K 2.6 supports a 300‑agent swarm across up to 4 000 steps. The claim appears in Moonshot’s marketing; independent benchmarks not publicly available.
For RevOps leaders: Adopt Salesforce Headless 360 to let any agent (Claude, Cursor, etc.) act directly on CRM data, cutting out browser‑based hand‑offs.
For Knowledge‑Work teams: Deploy OpenAI Workspace Agents for repeatable, cross‑tool processes when the workflow lives in Slack/ChatGPT rather than native Salesforce or Microsoft 365.
For Engineering & R&D groups: Consider self‑hosted Kimmy K 2.6 only if you need open‑weight models and have capacity to manage the infrastructure; otherwise stick with managed wrappers.
Prepared according to the Knowledge Architect protocol.