Workspaces
Log in Sign up
← Briefing history

This cycle's headline is price compression from two directions at once: Google commoditized the "source-grounded notebook" at…

Read-only snapshot of Autonomous research competitive landscape

Watch activity
Sep 19, 2026 · 5 findings · ran 4m

Autonomous Research Tools: Market Map Digest

TL;DR

This cycle's headline is price compression from two directions at once: Google commoditized the "source-grounded notebook" at $0–$20/month, and Moonshot shipped a full vertical research product for finance at $7–$104/month — an order of magnitude under Western enterprise seats. Meanwhile Elicit is running the dual-layer strategy (destination workspace and MCP evidence backend) as a live proof of concept, which is exactly the fork Hey, Lefty has to choose about.

The bottom of the market just collapsed under Google's free tier

Google's rebrand of NotebookLM into "Gemini Notebook" (July 2026) turns source-grounded research from a differentiator into a bundled subscription feature. Google's own positioning claim: "As a research assistant, Gemini Notebook's advantage is being source-grounded. This helps deliver more accurate answers and insights based on your actual material — reducing the likelihood of AI errors and hallucinations" (Gemini Notebook plans page). The tier ladder rides Google AI subscriptions — Plus $4.99/mo, Pro $19.99/mo, Ultra up to $199.99/mo — and the notebook itself "cannot be bought as a standalone product at any price" (FelloAI).

The enterprise SKU is the real price anchor: NotebookLM Enterprise starts "at roughly $9 per license per month with a 15-license minimum … the only path that adds enterprise security controls like VPC Service Controls, IAM access management and audit logging" (FelloAI). For the market map, any workspace whose pitch reduces to "chat with your sources, get cited answers" is now competing against a $0–$20 bundle with a $9/seat secure tier — a startup at $30–50/seat needs capabilities Google doesn't ship, not parity features (gemini-notebook-rebrandfelloai.comnotebooklm.google).

What to watch: Whether the chat hosts (OpenAI, Anthropic, Google) ship a native research workspace inside the chat app itself — Gemini Notebook is still a separate product, so that tripwire remains untriggered (notebooklm-gemini-notebook-rebrandfelloai.comnotebooklm.google).

Elicit is testing both layers of the bifurcation at once — and winning customers on each

Elicit is now the clearest live test of the destination-vs-backend question from last cycle's digest. On the destination side, its September 1 multiplayer launch frames the ambition directly: "AI has enabled a single researcher to do more work than ever before. Today, we're launching a set of features that make Elicit more collaborative and give teams a shared environment where researchers and AI agents can work together" (Elicit changelog, Sept 1 2026). Skills, Collaborative Sessions, and Shared Projects make it a persistent team environment, not a chat wrapper.

On the backend side, its July API/MCP launch exposes three endpoints — Search (138M+ papers), Reports, and a Systematic Review flow where "each decision recorded, so the review is auditable end to end" — served over MCP so "Elicit can return cited answers directly in Claude, ChatGPT, Gemini, or any MCP-compatible client" (Elicit blog, Jul 15 2026). The customer pull is telling: a life-sciences investment firm cut ~20 hours of research per asset, and a VC fund's founder dismisses most AI research output as "vibes on the internet" — the same provenance complaint that anchored last cycle's trust discussion (elicit-research-agent-api-mcpelicit.comsupport.elicit.com).

Pricing sits at the premium band: free Basic, Pro at $49/user/mo, Scale at $169/user/mo, Enterprise custom — with the market-wide shift to monthly usage pools since June 30 (Elicit pricing). For Hey, Lefty, Elicit is both a candidate MCP source to call and a competitor for the workspace seat in science-adjacent research.

What to watch: Whether "Skills" — Elicit's reusable team research procedures, echoed by Moonshot's 9 financial Skills — hardens into the market's standard packaging unit for expertise, forcing every orchestrator to adopt the same vocabulary.

A frontier lab just ran the full vertical playbook at consumer prices

The sharpest new data point is Moonshot's September 17 launch of Kimi for Financial Services: proprietary data partnerships, packaged workflows, compliance controls, and named institutional clients — at 49–699 yuan per month. CNBC: "Subscriptions to Kimi start at 49 yuan ($7.31) a month, and can go up to 699 yuan ($104.23)" (CNBC, Sep 17 2026). That top tier undercuts Western enterprise research seats by roughly an order of magnitude versus Elicit's $169 Scale tier.

The architecture is the more strategically important detail, because it uses the same open primitives everyone else is converging on: "This data connects directly into research tasks via MCP; key figures can be checked against the original source, with provenance and methodology clearly traceable" (kimi.com announcement). Data partners include S&P Global Market Intelligence, Crunchbase, Wind, and direct EDGAR/IMF/World Bank/FRED access, with users named as CICC and Hong Shan (CNBC). When a frontier-model lab can assemble MCP data plumbing plus Skills plus a compliance gateway co-built with CITIC Construction Investment Securities at open-weight cost structure, "access to data" stops being a moat — differentiation moves entirely to evidence standards, audit, and orchestration quality.

What to watch: International expansion — if Kimi for Financial Services ships an English product or named non-Chinese clients, the $7–$104 price band lands directly on Sacra and the institutional MCP vendors' turf (kimi-financial-services-moonshotcnbc.comkimi.com).

What surprised us

  • The compliance features are arriving at commodity prices. Moonshot bundles five compliance measures — including mandatory human review and full audit traceability — into a $7–$104 product (kimi.com announcement), while Google sells audit logging at ~$9/seat (FelloAI). "Enterprise-grade governance" as a premium justification is dying fast.
  • Western data vendors are already inside a Chinese lab's product. S&P Global Market Intelligence and Crunchbase powering a Beijing frontier lab's finance tool is a quiet landmark — the data layer is for sale to anyone, which validates the MCP-orchestrator thesis and kills the "exclusive data partnership" moat story.
  • Elicit's benchmark self-report is bold but unverifiable. Its claim that Smartest mode covers 76.7% of key decision considerations vs. 68.8% for Claude Opus 5 Max comes from an internal benchmark (Elicit changelog) — a reminder that in this market, the vendor is also the referee, which is itself an argument for independent evaluation layers.
  • The free tier is doing the commoditizing, not the paid tiers. Google's free tier includes 10 Deep Research sessions/month and 100 notebooks (FelloAI) — enough to absorb most casual research demand before a single dollar changes hands.

Open threads worth a vote

Findings from this cycle

Current topic brief

Shown for context; the brief may have changed since this cycle ran.

What is the market for autonomous or AI research tools? There's gemini deep research, google scholar, perplexity. Sacra is another research platform. What do all of these tools do? What are their features? Their value prop? Their core technology? Their data and where does it come from? Who do they sell to? what is the pricing/business model? Help me build a market map to see where Hey, Lefty fits and we should position it.