💥 What Matters at Seed, Are… | Data Driven VC Summary

1. Key Themes

Theme 1: AI Agent Capabilities Are Accelerating Faster Than Previously Estimated — and Faster Than Fund Deployment Timelines

METR's updated research reveals that AI task-completion horizons are doubling nearly twice as fast as their prior estimate. This has immediate implications for how investors should think about the durability of agent-based startups.

"Post-2023 doubling time for AI task-completion horizons is 130.8 days (about 4.3 months), roughly 20% faster than METR's prior estimate. The 2024 to 2025 sub-period showed doubling every 4 months, with the trend accelerating rather than plateauing."

The author's takeaway sharpens the investment urgency:

"The agent thesis investment window is closing faster than typical fund deployment timelines and entry prices already reflect the acceleration."

And critically for founders building on current model limitations:

"Founders building 'AI does X for Y hours' should assume their moat from current model limitations has a 4-month half-life."

Theme 2: The Venture Liquidity Crisis Is Structural, Not Cyclical

The WEF/Stanford GSB report reveals a deeply stalled capital recycling mechanism across the VC ecosystem, with concentration intensifying at both the top and the tails.

"Roughly 1,920 VC-backed unicorns remain privately held globally, holding more than $7.3T in valuation and roughly $3T in unrealised NAV on fund balance sheets. Of those unicorns, 59% were founded more than 10 years ago and 20% more than 15 years ago."

DPI has materially declined for prime-vintage funds:

"Funds in their 5-to-10-year prime return window historically returned around 20% of their value to LPs. By end of 2025, that figure had fallen to 12%."

Capital concentration is compounding the problem:

"The 10 largest VC funds captured 42.9% of all capital raised in 2025. Median time to close a US VC fund stretched to a record 15.3 months, up from 9.7 months in 2022."

And the exit path for most portfolio companies is narrowing:

"Secondaries are now the most active liquidity lever at roughly 30% of US VC exit value, but the top 20 names absorb 86% of that volume, leaving the long tail without a real exit path."

Theme 3: The Pre-Seed Market Has Bifurcated — The Middle Is Disappearing

Carta's Q1 2026 State of Pre-Seed report (based on ~3,000 US rounds, $2.3B in capital) shows a structural split in the earliest stage of venture, with AI dominating the dollar flow.

"AI now takes 50% of pre-seed dollars, up from 30% [a few years ago]."

The middle of the market is being hollowed out:

"Rounds between $1M and $2.5M fell from 24% of pre-seed rounds in Q1 2023 to just 18% in Q1 2026. Sub-$1M rounds are growing in share while $2.5M+ rounds stay stable, leaving a bifurcated market."

The implications for traditional pre-seed funds are direct:

"Funds in that range face a choice between going smaller and earlier or moving up to compete on $2.5M+ rounds."

Theme 4: The End of SaaS as a Default Software Investment Thesis

Two separate sources — a16z and Point Nine's Christoph Janz — converge on the same structural shift: pure application-layer SaaS is losing its defensibility, and durable value is migrating elsewhere.

From the a16z GTM piece:

"The system-of-record era of GTM software is ending, and the next decade of enterprise value will accrue to the orchestration layer that reads and writes across the CRM rather than the CRM itself." "Salesforce at ~$140B vs HubSpot at ~$9B captured 20 years of GTM value... The thesis is that the database that produced that outcome is being demoted to infrastructure."

From Point Nine's portfolio evolution:

"When the canonical European B2B SaaS investor publicly says 2010-style SaaS investing is dead, it confirms what the a16z piece argues from the other direction: application-layer defensibility is collapsing, and the durable bets are migrating to physical, biological and foundation-model domains where AI plus capital intensity creates real moats." "The question at seed is now domain depth and proprietary data, not GTM motion. Generalist software theses are aging out faster than most LP decks acknowledge."

Theme 5: The GTM Software Layer Is Shifting From Seats to Outcomes — and Expanding Into the Labor Budget

a16z's thesis on the evolving GTM stack identifies a specific and measurable unlock: AI enabling software to capture a larger share of total GTM spend by substituting for human labor without reducing it.

"Historically GTM software has been 5 to 10% of total GTM spending versus 90%+ on payroll. The a16z thesis is that AI is the first wedge that lets software expand into the labor budget without cutting headcount."

The seat-to-API substitution pattern is the key signal:

"One customer cutting Salesforce seats from 10+ to 2 human seats and 1 API seat, while spend rose 83% from $12K to $22K per year. CRM usage has actually risen since AI tools were adopted at scale."

2. Contrarian Perspectives

Perspective 1: VC Track Record Persistence Is Weaker Than LPs Believe — Relationships Compensate for the Lack of Information

The conventional LP wisdom is to back proven GPs with demonstrated top-quartile returns. The Odin/Dan Gray data challenges that assumption directly.

"Top-quartile VC funds repeat top-quartile status roughly 45% of the time based on fully realized performance, but only 33% of the time based on data known at the time of investment."

Relationships are doing more work than diligence:

"An LP is 1.78x more likely to select an emerging manager if there is an existing personal relationship, versus only 1.21x for established GPs. That gap is the relationship compensating for uncertainty rather than information."

Implication: If persistence is only 33% at the point of decision and relationships dominate selection, LPs relying on pedigree and interim TVPI are systematically overpaying for false signal — and structurally underallocating to emerging managers where the actual return opportunity may be widening.

Perspective 2: High Founder NPS Is a Red Flag for LP Returns, Not a Feature

Counter to the prevailing "founder-friendly fund" narrative used heavily in GP marketing:

"Drawing on 20VC data, the highest-NPS managers tend to deliver lower DPI. Hands-off, founder-friendly behavior often correlates with weaker realized exits, because it can mean less monitoring and more tolerance for inflated marks."

Implication: The "founder-friendly" brand has become a fundraising tool that may inversely signal weaker governance and accountability — meaning LPs who weight founder NPS positively in GP selection may be selecting for lower actual returns.

Perspective 3: AI Is Growing CRM Usage, Not Killing It — the Threat Is to the Business Model, Not the Category

Conventional narrative: AI agents will replace CRMs. The a16z data suggests the opposite dynamic on engagement, while confirming a business model disruption.

"CRM usage has actually risen since AI tools were adopted at scale."

The disruption is at the monetization layer, not the usage layer:

"The unit of value migrates from seats to outcomes... The durable moat in GTM software shifts from data accumulation to orchestration across systems."

Implication: Incumbents like Salesforce face a monetization model threat, not a usage collapse — which is actually harder to defend against, because growth metrics can look healthy while the unit economics are being hollowed out underneath.

3. Companies Identified

Salesforce

Description: Enterprise CRM incumbent
Why mentioned: Used as the benchmark for 20 years of captured GTM software value, and as the case study for the seat-to-API substitution trend
Quote: "Salesforce at ~$140B vs HubSpot at ~$9B captured 20 years of GTM value... one customer cutting Salesforce seats from 10+ to 2 human seats and 1 API seat, while spend rose 83%."

HubSpot

Description: Mid-market CRM and marketing platform
Why mentioned: Cited alongside Salesforce as evidence of extreme value concentration in the prior SaaS era
Quote: "Salesforce at ~$140B vs HubSpot at ~$9B captured 20 years of GTM value."

OpenAI, Anthropic, Scale AI, xAI, Project Prometheus

Description: Leading AI foundation model and infrastructure companies
Why mentioned: Cited as evidence of extreme capital concentration within AI — five companies absorbing a disproportionate share of global VC
Quote: "OpenAI, Scale AI, Anthropic, Project Prometheus and xAI absorbing 20% of total global VC across just five rounds."

Sensmore

Description: Autonomous dump truck startup (200-ton vehicles for open-pit mines)
Why mentioned: Recent Point Nine portfolio addition, emblematic of the "bits to atoms" investment thesis shift
Quote: "Recent Point Nine investments include Sensmore (200-ton autonomous dump trucks for open-pit mines)."

Serova

Description: Personalized cancer vaccine company
Why mentioned: Point Nine portfolio, representing the move into biological/physical domain bets
Quote: "Serova (personalized cancer vaccines)."

EraDrive

Description: Autonomous navigation for spacecraft
Why mentioned: Point Nine portfolio, part of the physical-world AI investment shift
Quote: "EraDrive (autonomous navigation for spacecraft)."

Vercept

Description: Foundation model for computer use (acquired by Anthropic)
Why mentioned: Point Nine portfolio exit, validating the foundation-model software bet
Quote: "Vercept (foundation model for computer use, recently acquired by Anthropic)."

Poolside

Description: Foundation model for software agents
Why mentioned: Point Nine portfolio, cited as the type of pure-software bet still worth making
Quote: "Poolside (foundation model for software agents)."

Sereact

Description: Warehouse robotics — picks objects it has never seen before
Why mentioned: Point Nine portfolio, illustrating AI-enabled physical-world automation
Quote: "Sereact (warehouse robots picking objects they have never seen)."

Hula Earth

Description: On-device AI identifying nearly 10,000 animal species
Why mentioned: Point Nine portfolio, cited in the "AI for the world outside the office" theme
Quote: "Hula Earth (on-device AI identifying nearly 10,000 animal species)."

Rerun

Description: Data infrastructure for robotics and computer vision
Why mentioned: Point Nine portfolio; infrastructure play in the physical AI stack
Quote: "Rerun (data infrastructure for robotics and computer vision)."

Forithmus

Description: Foundation models for medical imaging
Why mentioned: Point Nine portfolio; represents domain-specific foundation model bet in life sciences
Quote: "Forithmus (foundation models for medical imaging)."

Zauber

Description: AI agents for sea and air freight forwarders
Why mentioned: Point Nine portfolio; vertical AI agent bet with deep domain specificity
Quote: "Zauber (AI agents for sea and air freight forwarders)."

4. People Identified

Andre Retterath

Description: Author of Data Driven VC newsletter; venture investor focused on data and AI-driven investment approaches
Why mentioned: Author and curator of the newsletter
Quote: "Hi, I'm Andre and welcome to my newsletter Data Driven VC which is all about becoming a better investor with data and AI."

Hamza Shad (Carta)

Description: Analyst at Carta
Why mentioned: Author of the State of Pre-Seed Q1 2026 report, based on ~3,000 US pre-seed rounds
Quote: "Hamza Shad at Carta released the State of Pre-Seed Q1 2026 report, based on roughly 3,000 US pre-seed rounds adding up to over $2.3B in cash raised in the quarter."

Christoph Janz (Point Nine)

Description: Co-founder and Managing Partner at Point Nine Capital; canonical European B2B SaaS investor
Why mentioned: Published a reflection on how his firm's investment thesis has shifted from SaaS to physical, biological, and foundation-model domains
Quote: "Christoph Janz at Point Nine published a reflection on how a B2B SaaS firm's portfolio has shifted, with recent bets sitting at the intersection of AI and either the engineered world or nature."

Dan Gray (Odin)

Description: Writer/analyst at Odin, an LP platform
Why mentioned: Author of a long-form essay on emerging manager selection, drawing data-driven parallels between LP-GP and VC-founder evaluation
Quote: "Dan Gray at Odin published a long-form essay on emerging manager selection, drawing parallels between LPs evaluating first-time GPs and early-stage VCs evaluating founders."

Steph Zhang, Gio Ahern, Alex Immerman (a16z)

Description: Partners/analysts at Andreessen Horowitz
Why mentioned: Authors of the a16z thesis arguing that the system-of-record era of GTM software is ending
Quote: "Steph Zhang, Gio Ahern and Alex Immerman at a16z argue that the system-of-record era of GTM software is ending."

Jason Lemkin (SaaStr)

Description: Founder of SaaStr; prominent SaaS investor and commentator
Why mentioned: Cited as the source for the Salesforce seat-reduction case study that anchors the a16z GTM thesis
Quote: "Citing Jason Lemkin at SaaStr, the piece notes one customer cutting Salesforce seats from 10+ to 2 human seats and 1 API seat, while spend rose 83%."

5. Operating Insights

Insight 1: Seed-Stage Founders Must Stop Treating Current Model Limitations as a Moat

The METR data makes a specific, actionable warning for founders building on top of current AI capability gaps:

"Founders building 'AI does X for Y hours' should assume their moat from current model limitations has a 4-month half-life."

The operating implication is that defensibility must come from proprietary data, domain depth, or workflow lock-in — not from being first to exploit what models cannot yet do. Any product strategy built on the assumption that today's capability ceiling holds for 12–18 months is likely already wrong.

Insight 2: In GTM Software, Screen for Outcome-Based Pricing Over Seat-Based Pricing

The a16z analysis identifies a specific, measurable signal that separates durable GTM AI companies from feature wrappers:

"The unit of value migrates from seats to outcomes... The clearest investor signal is the seat-to-API substitution at constant or rising spend, since it shows AI labor budget being unlocked without payroll erosion."

For operators, this means: if your pricing model is still seat-based and your product is displacing human workflows, you are likely underpriced and at risk of being repriced downward. The companies capturing durable GTM value will charge on pipeline generated, deals closed, or revenue influenced — not on user licenses.

Insight 3: LP and VC Selection Processes Are Both Distorted by Relationship Bias Over Actual Signal

For operators raising institutional capital (at both the fund and company level), the Odin data provides a tactical insight:

"An LP is 1.78x more likely to select an emerging manager if there is an existing personal relationship... That gap is the relationship compensating for uncertainty rather than information."

The practical takeaway for founders and emerging GPs alike: in the absence of hard performance data, relationship surface area substitutes for diligence. Investing in network density — not just pitch quality — is a high-return operating activity in early fundraising cycles.

6. Overlooked Insights

Insight 1: Domain Divergence in AI Capability Growth Is Wider Than the Headline Doubling Rate Suggests

The METR report's finding on divergence across domains is buried beneath the headline 4.3-month doubling figure, but is arguably more important for sector-specific investment decisions:

"Software and reasoning tasks have 50 to 200+ minute horizons doubling every 2 to 6 months. Visual computer use lags 40x to 100x behind on absolute horizon length. Self-driving improves at roughly 0.6 doublings per year, a fraction of the software pace."

This means the risk profile of an AI agent bet in software is categorically different from one in physical-world or visual domains — and that investors bundling "AI agents" into a single thesis are likely mispricing risk across the portfolio.

Insight 2: Convertible Notes Are Nearly Extinct at Pre-Seed — But Biotech, Energy, and Medical Devices Are the Exception

The Carta data reveals that SAFEs have almost entirely displaced convertible notes at the pre-seed stage, but the remaining note volume is clustered in specific sectors with structural reasons to prefer additional terms:

"Convertible notes hit a record low of just 7% of pre-seed rounds and 8% of pre-seed dollars in Q1 2026. The remaining convertible-note volume clusters in biotech, energy and medical devices, where investors still prefer the additional terms."

This is a quiet signal that deep-tech and life-science pre-seed deals are still operating under different structural norms than the broader market — which has implications for how to structure deals, what precedents apply, and which legal/term frameworks remain relevant in those sectors.