FoundersOperators, strategy & lessons
Dragonfly's Haseeb Qureshi says founders now send VCs a company.md file for AI screening agents
Some founders now send VCs a company.md, a markdown summary meant for AI screening agents, alongside the pitch deck, Dragonfly's Haseeb Qureshi wrote. A This Week in Startups panel weighed the practice, which echoes the llms.txt convention and has no published schema.
From the This Week in Startups video of November 12, 2026 and cited sources: Qureshi wrote that some founders now add a markdown file, company.md, to their deck and data room so that VCs' AI agents can screen the company. [1] [2]
SOURCES[1] Founders Are Now Sending VCs a File Written for AI youtube.com[2] TWiST newsletter: Optimize your pitch for the human… twistartups.substack.com
AT Pulse research deskPublished · Updated · 5 min read · AI-assisted analysisAI-assisted, not human-edited. Methodology Corrections Who owns this site
- What happened
- Qureshi wrote that some founders now add a markdown file, company.md, to their deck and data room so that VCs' AI agents can screen the company. [1] [2]
- Original source
- This Week in Startups video, independent perspective
- Brief by
- AT Pulse research desk AI-assisted; publisher and contact on the desk page
- When
- November 12, 2026 video published
- Covered
- Nov 12 edition
- Evidence
- 8 sources across 8 sites; transcript sourced
- 1Founders Are Now Sending VCs a File Written for AIyoutube.com
- 2TWiST newsletter: Optimize your pitch for the human…twistartups.substack.com
- 3Answer.AI: /llms.txt proposalanswer.ai
- 4llmstxt.org: The /llms.txt file, v2llmstxt.org
- 5micro1: Q3 update (Oct 12, 2026)micro1.ai
- 6FourWeekMBA: micro1 passes $1B run ratefourweekmba.com
RisingHow it’s measured
YouTube video · 4.5K views · 38 likes
Reach blends public engagement (views, likes, comments, shares, stars) from the platforms where the story or its original source lives: YouTube, Hacker News, GitHub, Substack and similar. A video and the article written from its transcript share one score, taken from the video. Platforms with no data are left out rather than counted as zero, and the age of a story never lowers its score.
Qureshi wrote that some founders now add a markdown file, company.md, to their deck and data room so that VCs' AI agents can screen the company. He called it a positive signal. On This Week in Startups, Jason Calacanis described investors questioning an LLM about the business instead of reading a deck or taking a call, and said he had not yet received one. Jenny Fielding of Everywhere Ventures welcomed it as the next step in agent-driven venture work. [1] [2]
There is no company.md standard. The closest precedent is llms.txt, a 2024 community proposal for a curated markdown summary that LLMs can read. It has a title, a short summary and sections that link to detail. Markdown also means a screening agent does not have to pull meaning out of PDFs, spreadsheets and chart images. [2] [3] [4]
AI deal scoring tends to hedge. One panelist said his fund's agents produce strong analysis and market maps. But on a 1-to-9 scale with no 5, they return a 3, 4, 6 or 7 almost every time, so the scores are low-conviction. He described no check of these scores against actual deal outcomes. [1]
Funds are also making themselves readable to AI agents. Fielding said Everywhere Ventures added FAQs and agent-friendly pages after an informal check found that almost none of the other VC websites it looked at were set up for agents. The firm is also spinning out Jelly, an in-house investor-data tool, as a cheaper alternative to PitchBook seats. [1]
So far the evidence is one investor's post and a few panel anecdotes. If more funds use AI agents to screen inbound pitches, a machine-readable summary could affect who gets a first meeting.
Read the full assessment
That would give founders a cheap advantage and give funds new claims to check.
Executive brief
A pitch-deck replacement now has a filename: company.md. Haseeb Qureshi, managing partner at crypto-focused VC Dragonfly, posted on X that some founders now send a markdown file written for AI to read, along with their deck and data room. He called it "a very positive signal" (TWiST newsletter). The reason is that more VCs hand their first screening of inbound pitches to AI agents. On This Week in Startups, Jason Calacanis said he had not yet seen one. Jenny Fielding of Everywhere Ventures welcomed the practice. Another panelist said AI deal scores are reliably "low conviction" (episode, 8:00). The evidence is anecdotal: no data on how widely this is used was found.
What changed and event timeline
Llms.txt proposed
Jeremy Howard of Answer.AI proposed a markdown file at a site's root that gives LLMs a curated summary of the site. It is the closest technical precedent for company.md ().
A VC fund rewrites its website for AI agents
Fielding says Everywhere Ventures added FAQs and agent-friendly pages after finding that about 99% of the VC sites it checked weren't built for AI agents. This is her own account, not an audit ().
Llms.txt v2
The specification was updated. It is still a community proposal, not a ratified standard ().
On or before 2026-11-11: Qureshi's company.md post
He wrote that founders send company.md to suit VCs' AI screening agents. The post had about 52.4K views and 784 likes, according to the newsletter recap ().
Panel debates "agentic venture."
The TWiST VC roundtable discussed company.md, AI-scored deals, a fund spinning out its own data product, and LPs shifting money into single-deal vehicles (SPVs) ().
Capabilities and access
- No model involved. company.md is a plain markdown file that any LLM can open.
- No standard exists. None of the reviewed sources contains a published company.md schema. Qureshi describes it as a convention, not a format.
- Related tool: Everywhere Ventures is opening a beta of "Jelly," an investor-intelligence tool it built in-house. Fielding says the waitlist is 1,000 people.
Read the full section
- No model involved. company.md is a plain markdown file that any LLM can open. Calacanis described VCs asking an LLM questions about the business instead of reading a deck or taking a call (8:18).
- No standard exists. None of the reviewed sources contains a published company.md schema. Qureshi describes it as a convention, not a format.
- Related tool: Everywhere Ventures is opening a beta of "Jelly," an investor-intelligence tool it built in-house. Fielding says the waitlist is 1,000 people. Pricing is undecided; the panel floated $30–50 a month (28:14).
Technical analysis for researchers and developers
- Same design as llms.txt: an H1 name, a blockquote summary, then sections linking to detail (llmstxt.org).
- Why markdown: it spares the screening agent from extracting meaning out of PDFs, spreadsheets and chart screenshots (TWiST newsletter).
- Evaluation behaviour: one panelist scores deals with several agents on a 1–9 scale with no 5. No validation against actual deal outcomes was described.
Read the full section
- Same design as llms.txt: an H1 name, a blockquote summary, then sections linking to detail (llmstxt.org). A company.md that follows this layout is easy for software to parse and keeps the context window small.
- Why markdown: it spares the screening agent from extracting meaning out of PDFs, spreadsheets and chart screenshots (TWiST newsletter).
- Evaluation behaviour: one panelist scores deals with several agents on a 1–9 scale with no 5. He says the AI returns 3, 4, 6 or 7 "almost 100% of the time": it hedges and avoids the extremes (14:07). No validation against actual deal outcomes was described.
Claims and evidence
- Founders are sending company.md: one investor's post, recapped by the TWiST newsletter. No independent count exists.
- About 99% of VC sites aren't agent-friendly: Fielding's informal check, no method described (8:54).
- AI scores bunch in the middle: one fund's internal experience (14:07).
Read the full section
- Founders are sending company.md: one investor's post, recapped by the TWiST newsletter. No independent count exists.
- About 99% of VC sites aren't agent-friendly: Fielding's informal check, no method described (8:54).
- AI scores bunch in the middle: one fund's internal experience (14:07).
- micro1, mentioned on the panel: Calacanis said his fund invested at a $12M valuation and that the company is at $1B revenue (6:35). He was talking up his own portfolio company. micro1 itself says it passed a $1B annual run rate and is profitable. Its Series B valued it at $4B, led by Dream Ventures with Microsoft, Javelin, Flybridge, G Squared, Up Partners and LG, plus a roughly $25M employee tender (micro1, Oct 12, 2026). These are company-announced, unaudited figures, and the round size was not disclosed (FourWeekMBA). Separate reports put the round at about $100M or more (Dealroom). That figure is not confirmed by the company.
Context and prior work
- llms.txt: the main precedent. OpenAI, Anthropic and Google publish llms.txt files for their developer docs, but no major AI provider has confirmed using the format in production (GreenGeeks).
- Algorithmic VC isn't new: the panel recalled Chamath Palihapitiya's "capital as a service," which funded companies based on their SaaS metrics (20:00).
Limitations, safety and contested findings
- A file written for an AI is also a way to steer that AI. Founders can frame claims for a model's reading.
- Picking versus supporting founders: panelists argued AI may analyse deals as well as, or better than, analysts.
- Low-conviction scores: the hedged AI scores above weaken the case for fully automated screening.
Read the full section
- A file written for an AI is also a way to steer that AI. Founders can frame claims for a model's reading. The reviewed sources don't discuss defences against this.
- Picking versus supporting founders: panelists argued AI may analyse deals as well as, or better than, analysts. They said trust and founder support stay human, e.g. being "first call on your worst day" (22:24).
- Low-conviction scores: the hedged AI scores above weaken the case for fully automated screening.
Business and practitioner implications
- Founders: a short, sourced company.md costs little and may get past AI screening. Put numbers in it that can be checked.
- Funds: making your own website and materials readable by AI agents may affect whether founders find you (8:54).
- Data vendors: small funds are building their own tools instead of paying for PitchBook, which Fielding said costs about $20K a seat plus $15K for API access (28:46).
Read the full section
- Founders: a short, sourced company.md costs little and may get past AI screening. Put numbers in it that can be checked.
- Funds: making your own website and materials readable by AI agents may affect whether founders find you (8:54).
- Data vendors: small funds are building their own tools instead of paying for PitchBook, which Fielding said costs about $20K a seat plus $15K for API access (28:46).
- LPs: panelists described LPs moving money from funds into SPVs, a trend partly driven by OpenAI and Anthropic rounds (40:14).
The source trail.
Sources (8)
Founders Are Now Sending VCs a File Written for AI
Transcript retrieved via youtube_auto_captions; language en. Timestamped text, not direct audiovisual review. Source: https://www.youtube.com/watch?v=IN-xNzbxN_M. Automatic captions/transcription may contain errors.
www.youtube.com