Market Map & Positioning: Where Does Hey, Lefty Fit?
Revised 2026-10-05: adds agentic.ai's October 2026 third-party rankings — an independent "Agenticness" 0–36 scoring of research agents — as an external cross-check on the map's layers and on the autonomy claims vendors make about themselves. Prior revisions added the academic 2×2 quadrant taxonomy, the free-incumbent retrieval layer (Scholar Labs), Consensus's agentic/MCP trajectory, and the emergent destination-vs-backend bifurcation.
A third-party autonomy scoring of the destination layer
Agentic.ai's "11 Best AI Research & Analysis Agents in 2026" (directory page updated October 2026) scores research tools on a 0–36 "Agenticness" scale — "our 0–36 measure of how independently a tool acts (capability, autonomy, planning, reliability, safety, and four more)." The rankings:
| Tool | Agenticness | Tier |
|---|---|---|
| ChatGPT Deep Research | 17/36 | Adaptive Collaborator |
| Agentic News | 14/36 | Guided Assistant |
| Gemini Deep Research | 11/36 | Guided Assistant |
| Undermind | 11/36 | Guided Assistant |
| Elicit | 10/36 | Guided Assistant |
| Claude Research | 10/36 | Guided Assistant |
| Perplexity Research | 9/36 | Guided Assistant |
| SciSpace | 8/36 | Guided Assistant |
| Scholarcy | 6/36 | Guided Assistant |
| Consensus | 4/36 | Reactive Tool |
| STORM by Stanford | 2/36 | Reactive Tool |
Three observations for the map:
- Nobody scores as fully autonomous. The best-scored destination tool (ChatGPT Deep Research, 17/36) tops out at "Adaptive Collaborator" — barely half the scale. The vendor marketing of Sept–Oct 2026 (dots, Automations, Computer) describes always-on agents; independent scoring still places even the leaders in guided-assistant territory. The gap between marketed autonomy and scored autonomy is itself a positioning opening.1
- The buyer's stated non-negotiable is verifiability, not autonomy. The guide's selection advice: "Citation quality is non-negotiable… If you can't verify an output, it's creative writing, not research."2 It also tells buyers to "evaluate the synthesis capability, not just the search" and to prefer "ongoing monitoring vs. one-time research." That matches the audit-grade vertical thesis (The Audit-Grade Vertical: Research Agents That Sell "Show Your Work" — Lumivero, Menos AI, HITZ) — show-your-work is the differentiator buyers are coached to demand.
- The list is destination-layer only. Every ranked tool is a destination product; the backend layer (Exa Agent Ultra, Parallel, the institutional MCP servers) is absent from this buyer-facing view3. The bifurcation (Destination vs. Backend: The Research-Tool Market Bifurcates — and Both Layers Converge on MCP) is real in procurement, but buyers are still shopping destinations — which is where an orchestrator's surface lives.
The guide's domain-first advice also maps cleanly onto the map's layers: "Academic research agents (Elicit, Undermind, Consensus) index peer-reviewed papers and clinical trials. News monitoring tools track real-time media sources. General research agents handle both but may lack depth in either. Pick the tool that covers where your knowledge lives."
-
An instance of Self-graded benchmarks mint a market for independent judges. — A third-party 0–36 autonomy scoring shows external measurement of vendor claims becoming a layer buyers consult directly. ↩︎
-
An instance of Verifiable provenance becomes the product once generation is cheap. — Buyer guidance makes verifiability — not autonomy or raw capability — the non-negotiable purchase criterion. ↩︎
-
An instance of Deep research bifurcates into destination products and per-request backends — survivors straddle both. — The bifurcation is real in procurement even while buyers still shop the destination layer. ↩︎