✍️Takeaways from DDVC x Slush Builders Breakfast 2026
1. Key Themes
Theme 1: Buy third-party data, build your proprietary first-party data layer
Andre argues that the era of stitching together data providers in-house is over. The edge now comes from fusing commodity market data with a firm's unique internal data and taste.
- Earlybird learned the hard way that data plumbing consumed the team's time: "Quickly, we spent more time with data engineering, plumbing, and maintenance than with generating actual insights from the data."
- The market has matured: "the data landscape has matured and the value of building the third-party data collection inhouse diminished."
- His prescription: "I would buy the key databases for third-party data and focus my attention on the integration of unique first-party data such as pitch decks, transcripts, or investment memos."
- Alpha logic: "Not having access to the leading platforms is like negative alpha - you definitely miss out. The positive alpha is generated by mixing first-party data and adding your unique taste."
Theme 2: Deal triage as a three-layer funnel (hard filters → probability scoring → taste)
A practical architecture for cutting through deal-flow noise, ending in codified firm-level preference.
- Layer 1, hard filters: "Build a triage by starting with hard filters like geography, stage, company age or maximum funding raised, for example nothing older than 10 years or with more than $20M raised."
- Layer 2, scoring: "add a probability of success scoring based on public research. Claude is your friend here, just ask what factors predict success. Keep it simple to start with."
- Layer 3, taste: "You can present the same investment opportunity to 5 people in the same firm and most likely get different assessments. That's the beauty of investing."
- Codifying taste: "At Earlybird, we built a dating-app style swipe interface to learn individual preferences. Once codified, find the common denominator - this is your firm's taste."
Theme 3: The AI-native VC stack: buy the chat layer, build the context layer, buy the action layer
Framed via Yohei's "Agentic Fund" session, this is a three-tier architecture with explicit build-vs-buy calls.
- Tier 1: "Chat as the interface with text or voice, mostly Claude or ChatGPT, which remains highly interchangeable —> buy"
- Tier 2: "Git repo with firm context, harness, workflows, skills, MCP connectors, IDs, etc., that triggers a master agent, Yohei calls this the Chief of Staff —> build"
- Tier 3: "Systems of record / action with best-of-breed agents like Harmonic's Scout or outbound email agents —> buy"
- Design principle: "The master agent should not replicate what's already available from system of record / action providers."
- Where the firm's differentiation lives: "The firm owns what's worth looking for."
Theme 4: AI transformation is a cultural and ownership problem, not a technical one
Firms over-index on tooling and underestimate adoption dynamics.
- "Most funds start with the technical challenges - finding the right use cases, tools, workflows, and more. This is all solvable, no rocket science. For long, however, majority of firms underestimate that the actual challenge is cultural adoption."
- The fix is senior ownership: "AI adoption correlates with seniority of the owner. Ideally, a GP with access to manco-level budget is the owner."
- Failure mode: "When the partners are still stuck in the old world but juniors are building, it'll most likely remain a nice-to-have but nothing critical to the firm."
2. Contrarian Perspectives
Don't build your own data collection, even if you're a data-driven fund
Conventional wisdom in data-driven VC has been that proprietary sourcing pipelines are a moat. Andre, whose firm spent a decade building one, reverses course.
- Evidence from experience: the team patched together "Crunchbase, Dealroom, Pitchbook, CB Insights, and more," then added "LinkedIn, Github, HackerNews" after missing startups, and ended up consumed by maintenance.
- His conclusion: "the value of building the third-party data collection inhouse diminished." The moat shifts from collecting data to combining first-party context with taste.
The AI owner should be a GP who invests and builds, not a technical hire or junior
Many firms delegate AI initiatives to ops staff, junior investors, or dedicated engineers. Andre argues the opposite.
- "Ideally, a GP with access to manco-level budget is the owner. Even better, if this GP is an investor and has a technical background, someone who invests and builds."
- Reasoning: "solutions are tailored to actual investor needs and there is a 'leading by example' culture."
- Supporting evidence: "As highlighted in our 2026 DDVC Landscape report, AI adoption correlates with seniority of the owner."
Start simple on scoring rather than engineering a sophisticated model
Rather than building elaborate predictive models upfront, he advocates a lightweight start.
- "Keep it simple to start with, you'll have plenty of time to refine it later, across deterministic versus probabilistic approaches, and more."
3. Companies Identified
Harmonic
- Description: Startup discovery/data platform; co-host and sponsor of the event.
- Why mentioned: Earlybird migrated its third-party data stack to Harmonic; its Scout agent is the example best-of-breed system of action.
- Quotes: "I wrote about our benchmarks and the decision to migrate to Harmonic"; "The best investors use Scout to understand startups, the people behind them, and the markets they compete in."; "Scout finds matching companies and explains why they fit, drawing on Harmonic's startup data and public sources."
Earlybird
- Description: Andre's VC firm.
- Why mentioned: Case study for a decade-long digitization journey, data-stack evolution, and taste-codification.
- Quotes: "At Earlybird, we started our digitization journey about a decade ago."; "At Earlybird, we built a dating-app style swipe interface to learn individual preferences."
- Description: Law firm; event co-host.
- Why mentioned: Co-hosted the Builders Breakfast.
- Quotes: "co-hosted with our partners Harmonic and Goodwin."
- Description: CRM for relationship-driven investors.
- Why mentioned: Example system of record for checking existing relationships in the agent workflow.
- Quotes: "The orchestrating agent then checks the shortlist against your CRM system like Affinity"; "Affinity tells you who you already know."
Claude (Anthropic) / ChatGPT
- Description: LLM chat interfaces.
- Why mentioned: Interchangeable chat layer to "buy"; Claude suggested for deriving success-prediction factors.
- Quotes: "mostly Claude or ChatGPT, which remains highly interchangeable"; "Claude is your friend here, just ask what factors predict success."
Crunchbase, Dealroom, Pitchbook, CB Insights
- Description: Third-party startup data providers.
- Why mentioned: The patchwork of sources Earlybird originally combined.
- Quotes: "Crunchbase, Dealroom, Pitchbook, CB Insights, and more."
- Description: LP database tool (sponsor placement).
- Why mentioned: Promotional mention for finding LPs.
- Quotes: "Find new LPs in our database of 30k+ active family offices, fund of funds, pensions & more via Final Close"
4. People Identified
- Description: Author of Data Driven VC; at Earlybird.
- Why mentioned: Host and author of the takeaways.
- Quotes: "Below are my key takeaways, and the agent swarm that ties them together"
Yohei
- Description: Speaker in DDVC's "The Agentic Fund" session (surname not given; likely Yohei Nakajima).
- Why mentioned: Framed the three-level AI-native fund architecture.
- Quotes: "I think Yohei framed this nicely in our 'The Agentic Fund' session, splitting it up into three levels"; "Yohei calls this the Chief of Staff"
5. Operating Insights
Pick an owner with budget authority and investing credibility
Assign AI transformation to a senior GP with manco-level budget access, ideally technical. "The best hack to solve for this is the ownership of your AI transformation."
Put your firm's playbook in a Git repo that drives an orchestrating agent
Encode sourcing criteria (what "early stage" means, target categories, relationship checks) so agents execute your judgment. In the example, the agent "reads the sourcing workflow in your Git repo. It learns what early stage means for your fund, which categories you care about (e.g. evals, observability, and permissions), and how to check whether you already know a team."
Capture individual preferences, then extract the common denominator
Use a swipe-style interface to collect each investor's reactions and codify the overlap as firm taste. Start with simple hard filters (e.g., "nothing older than 10 years or with more than $20M raised") before layering scoring.
6. Overlooked Insights
Demand for this conversation is high, and the topics signal where small funds are stuck
With "more than 100 people on the waitlist" for 25 seats, the unanswered questions are telling: "Do we need a CRM? Build or buy?", "Can AI assess founders? Where does judgment matter?", and "How to make insights actionable?" These suggest that actionability and founder assessment remain unresolved even among senior fund leaders.
MCP is becoming the connective tissue between fund-internal agents and vendor tools
The worked example routes requests "to Harmonic's Scout via MCP," implying data and tooling vendors that expose MCP interfaces become easier to plug into an orchestrating agent, a potential selection criterion for buy decisions.