Agnost AI

Product analytics for AI agents to track intents, drop-offs, and retention.

Website: https://agnost.ai/

Cover Block

PUBLIC

Field Value
Name Agnost AI
Tagline Product analytics for AI agents to track intents, drop-offs, and retention
Headquarters San Francisco, United States
Founded 2024
Stage Pre-Seed
Business Model SaaS
Industry Developer tools / AI observability
Technology Type AI / Machine Learning
Geography North America
Growth Profile Venture Scale
Founding Team Co-Founders (2)
Funding Label Pre-Seed

Links

PUBLIC

Executive Summary

PUBLIC

Agnost AI is a San Francisco pre-seed company building product analytics specifically for AI agents and Model Context Protocol (MCP) servers, a category that did not meaningfully exist eighteen months ago [Agnost AI]. The company was founded in 2024 by Shubham Palriwala (Cofounder and CEO) and Shubham Patel, with Palriwala bringing roughly five years of backend engineering experience across Cisco, Formbricks, Onboarding.club and Google Season of Docs prior to starting the company [RocketReach; GetProg.ai]. The product surfaces conversation-level metrics, intent conversion, drop-off points, satisfaction signals and a proprietary Agent Experience Score, distributed via Python and JavaScript SDKs that target developers shipping AI-native applications [Agnost AI Docs; Agnost AI Blog]. Backers disclosed to date include Entrepreneurs First and Transpose Platform, both early-stage talent and infrastructure investors, with round size and valuation not publicly disclosed. The business model is SaaS aimed at developer teams building agents, MCP servers and plugins, a buyer profile that overlaps with the customer bases of Datadog, PostHog and LangSmith but with a narrower wedge focused on agent behavior rather than infrastructure telemetry. Over the next twelve to eighteen months the questions worth tracking are whether MCP adoption broadens beyond the Anthropic developer community, whether Agnost can convert its content footprint into paying logos, and whether the team layers on a sales motion before larger observability incumbents extend natively into agent analytics.

Data Accuracy: YELLOW -- Founder identities and product scope confirmed by company sources, LinkedIn and RocketReach; funding details remain undisclosed.

Taxonomy Snapshot

Axis Value
Stage Pre-Seed
Business Model SaaS
Industry / Vertical Developer tools, AI observability
Technology Type AI / Machine Learning, analytics SDKs
Geography North America (San Francisco)
Growth Profile Venture Scale
Founding Team Two technical co-founders
Funding Pre-Seed, amount undisclosed

Company Overview

PUBLIC

Agnost AI was incorporated in 2024 and operates out of San Francisco, with its X account joined in October 2024 marking the earliest public footprint of the brand [X, October 2024]. The founding thesis, as expressed across the company blog, is that conventional product analytics tools (Mixpanel, Amplitude, PostHog) were designed for click-and-pageview funnels and do not capture the unit of value in agent products, which is the conversation and the intent it resolves [Agnost AI Blog]. The company positions itself as the analytics layer for AI-native products, with native instrumentation for MCP servers, plugins and tool calls.

The two co-founders are both named Shubham. Shubham Palriwala serves as Cofounder and CEO, a role confirmed both on LinkedIn and in a recorded podcast appearance discussing MCP server metrics [LinkedIn; YouTube]. Shubham Patel, who previously studied at the University of Texas at Dallas, is the second co-founder [LinkedIn]. Investors disclosed publicly are Entrepreneurs First, the talent investor known for forming founding teams pre-idea, and Transpose Platform, an early-stage AI infrastructure investor.

Key milestones in the public record are sparse and consistent with a pre-seed timeline: company formation and brand launch in 2024, publication of the Python SDK and MCP-focused npm package (current version 0.1.10, last published roughly a month before the date of this report) [npm], a steady cadence of technical blog posts on MCP analytics, and at least two podcast appearances by Palriwala on the economics and metrics of MCP servers [YouTube]. No customer logos, revenue figures or headcount have been publicly disclosed.

Data Accuracy: YELLOW -- Founding year, location and founders cross-confirmed; legal entity, exact incorporation date and milestone dates beyond 2024 are not publicly available.

Product and Technology

MIXED

The product is an analytics platform purpose-built for AI agents and MCP servers. According to the company's own documentation, the core surface is a Conversation SDK that lets developers "track AI interactions, monitor performance, and analyze conversations in your Python application" [PUBLIC] [Agnost AI Docs]. A JavaScript/Node distribution is published as the agnost npm package, described as an "Analytics SDK for Model Context Protocol Servers," currently at version 0.1.10 with nine dependent projects in the public registry [PUBLIC] [npm]. The instrumentation captures conversation-level events, tool and plugin invocations, MCP task lifecycles, and what the company calls satisfaction signals (explicit and implicit user feedback within an agent interaction) [PUBLIC] [Agnost AI Blog].

On top of the raw event stream, Agnost surfaces several derived constructs that constitute its product opinion. The Agent Experience Score is presented as a single composite number for agent quality, with the company benchmarking customer support agents in the 72-80 range and noting that teams with intent-specific routing tend to cluster in the top quartile [PUBLIC] [Agnost AI Blog]. Other named metrics include an Intent Resolution Rate proxy, a Frustration Index, Health Scoring and per-customer cost attribution for Apify Actors, the latter aimed at giving agent operators a usage-and-cost view per end customer [PUBLIC] [Agnost AI Docs; Agnost AI Blog]. The blog also discusses long-running MCP task patterns under SEP-1686, suggesting the product is being shaped against the most current MCP protocol drafts rather than a static feature set [PUBLIC] [Agnost AI Blog].

The technology stack is not formally disclosed. SDK distribution channels (PyPI-style Python package and the public npm registry) and a developer-docs-first go-to-market suggest a self-serve, bottoms-up motion typical of modern observability tools (inferred from public artifacts). No information is published about backend storage, query engine or hosting model, and there are no public security certifications listed at the time of writing. Buyers requiring SOC 2 or comparable attestation should request status directly.

Data Accuracy: GREEN -- Product surface, SDK availability and named metrics confirmed by Agnost AI's documentation, blog and the npm registry.

Market Research and Opportunity

PUBLIC

Agent analytics is emerging as a distinct subcategory inside the broader AI observability and developer tools market, driven by the rapid adoption of LLM-based agents and the Model Context Protocol.

No third-party market sizing report dedicated specifically to "AI agent analytics" appears in the cited research, so any TAM figure for that exact wedge would be speculative. The honest framing is to triangulate from analogous categories. Product analytics broadly (Amplitude, Mixpanel, PostHog, Heap) is an established multi-billion-dollar software category, and AI/LLM observability (Datadog's LLM Observability product, Arize, LangSmith, Helicone, Braintrust) has materialised in the past 24 months as a parallel category aimed at model and prompt monitoring. Agnost positions itself at the intersection: not infrastructure telemetry, not model evaluation, but conversation-and-intent analytics for shipped agent products.

The demand drivers visible in the cited research are concrete. First, the Model Context Protocol is creating a new server type (MCP servers) whose operators need usage metrics; Agnost's own writing on "how to get more usage on your MCP server" and SEP-1686 long-running tasks is aimed squarely at that audience [Agnost AI Blog]. Second, the company argues in its blog that AI products fail on dimensions traditional analytics cannot see, namely "resolution without satisfaction," where an agent technically completes a task but degrades the user experience [Agnost AI Blog]. Third, customer support is cited as the most mature agent vertical, with benchmarks of 72-80 on the Agent Experience Score and a clear top quartile for teams using intent-specific routing [Agnost AI Blog]. These are signals that buyers in support-agent and MCP-server categories are already instrumenting and comparing.

Adjacent and substitute markets matter to the bull case. The most direct substitutes are general LLM observability tools (LangSmith, Arize, Helicone) extending upward into product metrics, and incumbent product analytics tools (PostHog, Amplitude) extending downward into AI events. Regulatory tailwinds are mild but real: enterprise buyers increasingly want auditable conversation logs for AI deployments, which any conversation analytics product is well-positioned to serve. Macro headwinds include the saturation of "AI tooling" budgets at many startups and the risk that hyperscaler-bundled telemetry (OpenAI usage dashboards, Anthropic console, AWS Bedrock metrics) commoditises basic agent metrics.

Reference benchmark Value Source
Agent Experience Score range, customer support agents 72-80 [Agnost AI Blog]
Public dependents on agnost npm package 9 [npm]
Current agnost npm version 0.1.10 [npm]

Analyst takeaway: the only hard numbers in the public record are product-level (benchmark ranges and SDK adoption signals), not market-level. The investable question is whether agent analytics becomes a standalone category like product analytics did, or a feature of LLM observability suites; Agnost is betting on the former.

Data Accuracy: YELLOW -- Category signals and product benchmarks sourced from Agnost AI's blog and the npm registry; no independent third-party market sizing for agent analytics specifically was located.

Competitive Landscape

MIXED

Agnost is positioned as a focused agent-and-MCP analytics specialist competing against both general-purpose LLM observability platforms and broader product analytics incumbents that are extending into AI events.

The segment map has three layers. The first layer is LLM and agent observability platforms whose primary persona is the AI engineer: LangSmith (LangChain), Arize AI, Helicone, Braintrust and Langfuse. These tools are generally strong on traces, evals, prompt versioning and cost monitoring, and they have raised meaningful capital and accumulated logos over the past two years. The second layer is incumbent product analytics: PostHog, Amplitude, Mixpanel and Heap. These tools dominate the PM and growth persona inside SaaS companies and are beginning to add LLM-event support. The third layer is platform-native telemetry: the dashboards shipped by OpenAI, Anthropic, AWS Bedrock and Google Vertex, which are free and good enough for basic usage tracking. Agnost's wedge sits between layers one and two: deeper into conversation and intent analytics than infrastructure-focused observability, but more AI-native than retrofitted product analytics [PUBLIC] [Agnost AI Blog].

Where Agnost has a defensible edge today, the evidence points to two areas. The first is native MCP support; the company appears to have been one of the earlier movers to publish an analytics SDK explicitly branded for MCP servers and to write technical content on MCP-specific patterns such as long-running tasks and plugin instrumentation [PUBLIC] [Agnost AI Blog; npm]. If MCP becomes a durable interoperability standard for agents, that early specialisation matters. The second is a clearly opinionated metric layer (Agent Experience Score, Frustration Index, satisfaction signals), which gives the product a point of view rather than a generic dashboard. Both edges are perishable: MCP-native support can be matched by any incumbent in a quarter, and metric definitions are easy to imitate but hard to make canonical without distribution.

Where Agnost is most exposed: distribution and brand. LangSmith ships with the most-used agent framework in the market and inherits its developer mindshare; PostHog has a multi-million-developer install base and an aggressive open-source motion; Datadog owns the enterprise observability budget. Agnost cannot match any of those distribution channels at pre-seed scale. The company also does not yet appear to own a specific vertical channel (for example, an exclusive integration with a major agent platform or a category-defining customer reference).

The most plausible 18-month scenario: winner if MCP adoption broadens beyond Anthropic's immediate ecosystem and Agnost becomes the default analytics SDK referenced in MCP server tutorials and starter kits, giving it a wedge similar to how Segment captured event collection a decade ago. Loser if PostHog or LangSmith ship a credible "AI product analytics" surface bundled into their existing free tier before Agnost has signed a meaningful base of paying logos; in that case the wedge collapses into a feature.

Opportunity

PUBLIC

The size of the prize is meaningful: if agent analytics becomes its own category the way product analytics did, the default SDK could become foundational developer infrastructure.

The headline opportunity

The single largest outcome Agnost could plausibly become is the default analytics layer for AI agents and MCP servers, in the same way Mixpanel and later Amplitude became the default for web and mobile product analytics, and the way Segment became the default event pipe. The cited evidence makes this reachable rather than aspirational on three specific points: (1) the category is genuinely new, with the Model Context Protocol only standardised recently and operators publicly asking how to instrument their servers [Agnost AI Blog]; (2) Agnost has already shipped working SDKs in both Python and JavaScript with public distribution and a small dependent base on npm [npm]; and (3) the company has staked out an opinionated metric vocabulary (Agent Experience Score, satisfaction signals, Frustration Index) that can become the lingua franca if it spreads through developer content [Agnost AI Blog]. None of these guarantees the outcome, but they are the prerequisites for category leadership.

Growth scenarios

Scenario What happens Catalyst Why it's plausible
MCP standard rides Agnost becomes the reference analytics SDK in MCP server tutorials, starter kits and protocol documentation Inclusion in widely shared MCP server templates and continued protocol momentum [Agnost AI Blog] Company is publishing protocol-level technical content (SEP-1686, long-running tasks) and already ships an MCP-branded npm package [Agnost AI Blog; npm]
Support-agent vertical wedge Agnost becomes the benchmarking standard for customer support agents, with the Agent Experience Score adopted as a vendor-comparison metric A flagship support-agent customer publishes results using Agnost's benchmarks (72-80 range) [Agnost AI Blog] Company already publishes vertical-specific benchmarks and identifies support as the most mature agent category [Agnost AI Blog]
Embedded analytics for agent platforms Agnost is bundled as the analytics layer inside one or more vertical agent platforms (voice agents, coding agents, RPA-style agents) Partnership or OEM deal with an agent-building platform SDK-first distribution and per-customer cost attribution feature in the Apify integration are consistent with an embed motion [Agnost AI Docs]

What compounding looks like

The flywheel for an analytics SDK is well understood: developer adoption of the SDK creates an event corpus, the event corpus enables benchmarking and comparative metrics that no individual customer can produce alone, and those benchmarks become a content and sales engine that pulls in the next cohort of developers. There is early evidence the loop has started turning: nine public dependents on the agnost npm package, a published benchmark range (72-80) for support agents derived from observed deployments, and a content cadence on the blog covering MCP, plugins and agent-experience patterns [npm; Agnost AI Blog]. Each new customer makes the benchmarks more credible, and more credible benchmarks make the next customer easier to land.

The size of the win

Comparable public-market reference points exist. Datadog has built a multi-tens-of-billions market capitalisation as the default observability layer for cloud applications, and Amplitude and similar product analytics players reached multi-billion outcomes serving the analogous need for web and mobile. If agent analytics becomes a standalone category at even a fraction of the scale of product analytics, the category leader could plausibly reach a unicorn outcome within five to seven years (scenario, not a forecast). At pre-seed and with no disclosed revenue, this is upside framing rather than projection; the realistic near-term win to track is whether Agnost converts its content-and-SDK footprint into a meaningful base of paying logos before incumbents close the wedge.

Data Accuracy: YELLOW -- Scenarios grounded in cited product capabilities and category signals from Agnost AI's own materials and the npm registry; market-cap comparisons are illustrative rather than sourced from a named third-party report.

Sources

PUBLIC

  1. [Agnost AI] Agnost: Product Analytics for AI Agents | https://agnost.ai/

  2. [Agnost AI Docs] Getting Started - Agnost AI | https://docs.agnost.ai

  3. [Agnost AI Blog] Blog index | https://agnost.ai/blog/

  4. [Agnost AI Blog] Long Running Tasks in MCP: The Call-Now, Fetch-Later Pattern | https://agnost.ai/blog/long-running-tasks-mcp/

  5. [Agnost AI Blog] How to Build Your Own Claude Code Plugin | https://agnost.ai/blog/claude-code-plugins-guide/

  6. [Agnost AI Blog] Resolution Without Satisfaction: Agent Experience | https://agnost.ai/blog/resolution-without-satisfaction-agent-experience/

  7. [Agnost AI Blog] Agent Experience Score | https://agnost.ai/blog/agent-experience-score/

  8. [Agnost AI Blog] How to Improve Your MCP Server | https://agnost.ai/blog/mcp-analytics-guide/

  9. [Agnost AI Blog] How to Get More Usage on Your MCP Server | https://agnost.ai/blog/increase-mcp-server-usage/

  10. [Agnost AI Blog] The 6 Metrics Every AI-Native Product Should Track | https://agnost.ai/blog/6-metrics-every-ai-native-product-should-track/

  11. [Agnost AI Docs] Apify Actors integration | https://docs.agnost.ai/apify

  12. [Agnost AI Docs] Roadmap | https://docs.agnost.ai/roadmap

  13. [npm] agnost package page | https://www.npmjs.com/package/agnost

  14. [X, October 2024] Agnost AI (@agnostai) | https://x.com/agnostai

  15. [LinkedIn] Shubham Palriwala profile | https://www.linkedin.com/in/shubhampalriwala/

  16. [LinkedIn] Shubham Patel profile | https://www.linkedin.com/in/shubham-patel-80203a219/

  17. [GetProg.ai] Shubham Palriwala, Co-Founder at Agnost AI | https://www.getprog.ai/profile/55556994

  18. [RocketReach] Shubham Palriwala, Agnost AI Co-Founder | https://rocketreach.co/shubham-palriwala-email_628647369

  19. [PitchBook] Agnost AI company profile | https://pitchbook.com/profiles/company/1258139-80

  20. [AIToolBook] Agnost Review | https://aitoolbook.ai/ai/agnost

  21. [YouTube] The Hidden Metrics Behind Successful MCP Servers, with Shubham Palriwala | https://www.youtube.com/watch?v=9BXrlbFlAQs

  22. [YouTube] MCP Analytics Q&A with Shinzo Labs and Agnost AI | https://www.youtube.com/watch?v=JVpbUbS73U4

  23. [Weekday] Shubham Palriwala profile | https://www.weekday.works/people/shubham-palriwala-shubhampalriwala

Articles about Agnost AI

View on Startuply.vc