Where AI capability ends and human judgment begins — a living record.
This document tracks what AI can and cannot reliably do in venture capital analysis. It reflects the state of the art across all AI tools on the market — not only what claude-vc provides.
Scope of claude-vc
Claude-vc is an open-source Claude Code skill plugin focused on deal screening, investment memos, cap table modeling, term sheet analysis, financial modeling, and KPI reporting. It does not intend to be an all-encompassing VC platform. It will gradually expand, but it does not aim to replicate proprietary tools like Harmonic (deal sourcing), Affinity (CRM), PitchBook (data), Luminance (legal review), or Standard Metrics (portfolio monitoring).
Achieving state-of-the-art performance in venture capital AI today requires integrating a dozen or more MCP servers, proprietary data subscriptions, and specialized platforms. Without a larger maintainer community, it is not feasible for a single open-source plugin to maintain those integrations. This document therefore serves as a map — it tells you what is possible with the right combination of tools, even where claude-vc does not yet cover it.
Status key
- Can Do: AI handles this reliably today (human quick review, not a rewrite)
- WIP: AI can attempt this, but outputs need substantive human rework
- Cannot (capability): No viable AI path today
- Cannot (regulatory): Blocked by licensing or institutional requirements, not AI capability
- Enabler1: The AI advancement or product that made this possible
Last reviewed: 2026-04-18
Deal Sourcing & Outreach
| Capability | Can Do | WIP | Cannot | Enabler |
|---|---|---|---|---|
| Inbound pitch triage and initial scoring | ✓ | Extended thinking, Harmonic signal detection | ||
| New investment opportunity identification | ✓ | Harmonic, Grata, EQT Motherbrain | ||
| Founder relationship tracking | ✓ | Affinity AI, 4Degrees | ||
| Co-investor and syndicate coordination | ✓ | CRM integrations, email MCPs | ||
| Back-channel reference calls | ✓ | Needs real-world interaction | ||
| LP and stakeholder communications | ✓ | LLM drafting, email MCPs |
Market Research
| Capability | Can Do | WIP | Cannot | Enabler |
|---|---|---|---|---|
| Industry landscape synthesis from public sources | ✓ | Web search, Perplexity, Claude | ||
| Market sizing (TAM/SAM/SOM) from pitch materials | ✓ | PDF vision, extended thinking | ||
| Competitive landscape mapping from web research | ✓ | Web search tool use | ||
| Sector trend analysis from public data | ✓ | Web search, long-context reasoning | ||
| Regulatory environment scanning | ✓ | Web search, legal corpus MCPs | ||
| Real-time market data and indices | ✓ | Bloomberg MCP, Refinitiv MCP | ||
| Primary market research (surveys, interviews) | ✓ | DiligenceSquared AI voice agents | ||
| Emerging market and whitespace identification | ✓ | Harmonic signals, Grata agentic search |
Company Research
| Capability | Can Do | WIP | Cannot | Enabler |
|---|---|---|---|---|
| Pitch deck data extraction (PDF) | ✓ | PDF vision, multimodal understanding | ||
| Public company profiling from web sources | ✓ | Web search tool use | ||
| Report generation (DOCX and markdown) | ✓ | Native file generation | ||
| Private company data access | ✓ | PitchBook Navigator, Crunchbase, Grata | ||
| Deal screening with structured scoring (0-100) | ✓ | Parallel subagents, extended thinking | ||
| Investment memo generation (10-section format) | ✓ | Parallel subagents, long-context generation | ||
| KPI benchmarking by auto-detected company type | ✓ | Python tool use, extended thinking | ||
| Comparable company analysis with market data | ✓ | PitchBook + Perplexity MCP | ||
| Factual claim verification against primary sources | ✓ | Web search with source citations | ||
| Systematic data room cross-referencing | ✓ | 1M context, self-verification |
Product Assessment
| Capability | Can Do | WIP | Cannot | Enabler |
|---|---|---|---|---|
| Product claims extraction from pitch materials | ✓ | PDF vision, multimodal understanding | ||
| Feature set and roadmap summarization | ✓ | Long-context reasoning | ||
| Technical architecture assessment | ✓ | Code analysis tools, computer use | ||
| User retention and engagement pattern analysis | ✓ | Analytics platform MCPs | ||
| Product-market fit validation | ✓ | Needs usage data access, user research | ||
| Hands-on product testing and UX evaluation | ✓ | Needs computer use at scale |
Financial Analysis
| Capability | Can Do | WIP | Cannot | Enabler |
|---|---|---|---|---|
| Burn rate and runway analysis | ✓ | Python tool use | ||
| 3-statement model generation (P&L, BS, CF) | ✓ | Python tool use, self-verification | ||
| Unit economics computation (LTV, CAC, payback) | ✓ | Python tool use, self-verification | ||
| Revenue projections (3-5 year forward) | ✓ | Python tool use, self-verification | ||
| KPI auto-detection and health assessment | ✓ | Extended thinking, self-verification | ||
| Cohort and retention curve analysis | ✓ | Analytics MCPs, Python tool use | ||
| Bulk portfolio-wide financial analysis | ✓ | ChatFin, Chronograph, 1M context | ||
| Audit-grade financial statements | ✓ | Regulatory: formal verification |
Valuation
| Capability | Can Do | WIP | Cannot | Enabler |
|---|---|---|---|---|
| Pre/post-money round modeling | ✓ | Carta, Pulley, Python tool use | ||
| Multiples-based valuation with industry ranges | ✓ | Python tool use, web search | ||
| DCF analysis from user-provided assumptions | ✓ | Python tool use, self-verification | ||
| Comparable company analysis with live data | ✓ | PitchBook + Perplexity MCP | ||
| Precedent transaction analysis | ✓ | Grata, PitchBook + Perplexity MCP | ||
| Conviction weighting on valuation outputs | ✓ | Needs calibrated confidence, human judgment |
Deal Structuring & Negotiation
| Capability | Can Do | WIP | Cannot | Enabler |
|---|---|---|---|---|
| Cap table modeling and dilution analysis | ✓ | Carta, Pulley, Python tool use | ||
| SAFE and convertible note conversion | ✓ | Carta, Pulley, OCF standard | ||
| Multi-series liquidation waterfall | ✓ | Carta, Pulley, Eqvista | ||
| Exit scenario modeling at multiple valuations | ✓ | Cap table platforms, Python tool use | ||
| Term sheet red-flag identification (NVCA baseline) | ✓ | Spellbook (10M+ contracts reviewed) | ||
| Cap table platform sync (Carta, Pulley) via MCP | ✓ | Cap table platform MCPs | ||
| Negotiation strategy and counter-offer structuring | ✓ | Needs relationship context, game theory | ||
| Binding legal document generation | ✓ | Regulatory: legal licensing required |
Technical Due Diligence
| Capability | Can Do | WIP | Cannot | Enabler |
|---|---|---|---|---|
| Technical claims summarization from materials | ✓ | Long-context reasoning | ||
| Technology stack identification | ✓ | Web search, code analysis tool use | ||
| IP and patent strength evaluation | ✓ | IPRally, Patsnap, Patlytics | ||
| Codebase quality and architecture review | ✓ | Code analysis agents, computer use | ||
| Scalability and infrastructure assessment | ✓ | Needs system access, load testing | ||
| Technical team capability evaluation | ✓ | Needs real-world interaction |
Legal Due Diligence
| Capability | Can Do | WIP | Cannot | Enabler |
|---|---|---|---|---|
| Common provision pattern flagging | ✓ | Spellbook, Luminance | ||
| NVCA baseline term comparison | ✓ | Spellbook VC clause library | ||
| Contract review (SHA, IP assignments, employment) | ✓ | Luminance, Kira/Litera (64% Am Law 100) | ||
| Regulatory compliance analysis | ✓ | Emerging legal AI tools | ||
| Legal opinions and formal advice | ✓ | Regulatory: legal licensing required | ||
| Cross-jurisdiction tax and structuring | ✓ | Regulatory: licensing + tax law MCPs |
Financial Due Diligence
| Capability | Can Do | WIP | Cannot | Enabler |
|---|---|---|---|---|
| Individual financial document analysis | ✓ | Hebbia Matrix, PDF vision | ||
| Private company financial data | ✓ | PitchBook, Morningstar, Grata | ||
| Financial model internal consistency checks | ✓ | Python tool use, self-verification | ||
| Portfolio company data connectivity | ✓ | Standard Metrics, Chronograph MCPs | ||
| Data room systematic cross-referencing | ✓ | 1M context, self-verification | ||
| Historical financials verification | ✓ | Needs SEC EDGAR MCP, audit trail access | ||
| Tax and transfer pricing analysis | ✓ | Regulatory: licensing + tax law MCPs |
Portfolio Management
| Capability | Can Do | WIP | Cannot | Enabler |
|---|---|---|---|---|
| CRM and deal pipeline tracking | ✓ | Affinity AI (auto-capture, relationship intel) | ||
| One-time portfolio summary generation | ✓ | Extended thinking, structured output | ||
| Board deck preparation assistance | ✓ | Long-context generation, native file output | ||
| Ongoing portfolio monitoring | ✓ | Standard Metrics, Chronograph, ChatFin | ||
| Scheduled recurring reports | ✓ | Standard Metrics, Visible.vc automation | ||
| Anomaly detection and alerts | ✓ | ChatFin anomaly engine | ||
| LP reporting preparation | ✓ | claude-vc portfolio skill, Standard Metrics |
Investment Decision & Closing
| Capability | Can Do | WIP | Cannot | Enabler |
|---|---|---|---|---|
| IC preparation materials | ✓ | Extended thinking, long-context generation | ||
| Decision framework structuring | ✓ | Extended thinking, structured output | ||
| Invest/pass recommendation | ✓ | Needs calibrated confidence, human judgment | ||
| Founder and team qualitative assessment | ✓ | Needs real-world interaction | ||
| IC facilitation and voting | ✓ | Needs multi-user collaboration | ||
| Closing coordination and execution | ✓ | Needs legal tooling, payment systems | ||
| Post-closing deliverable management | ✓ | Workflow automation, persistent agents |
Changelog
Dates reflect when the enabling product shipped, not when adoption matured.
| Date | Change | Capability affected | Direction | Enabler |
|---|---|---|---|---|
| 2022-09 | AI contract drafting and review | Common provision pattern flagging | Human -> WIP | Spellbook launch |
| 2022-11 | AI-powered deal sourcing signals | New investment opportunity identification | Human -> WIP | Harmonic Series A, early product |
| 2022-11 | Basic memo drafting and market summaries | Investment memo generation | Human -> WIP | ChatGPT (GPT-3.5) |
| 2023-03 | Professional-grade investment analysis | Industry landscape synthesis | WIP -> Can Do | GPT-4 reasoning quality |
| 2023-06 | Structured API tool calling for data access | KPI benchmarking, financial model checks | Human -> WIP | OpenAI function calling |
| 2023-07 | Full document analysis (100K context) | Pitch deck data extraction | Human -> Can Do | Claude 2 |
| 2023-09 | AI relationship intelligence in CRM | Founder relationship tracking | Human -> WIP | Affinity AI features |
| 2023-11 | Reliable structured data extraction | KPI benchmarking, deal screening | WIP -> Can Do | GPT-4 Turbo, JSON mode, 128K context |
| 2024-03 | Pitch deck and chart visual analysis | Pitch deck data extraction, product claims | Human -> Can Do | Claude 3 vision + 200K context |
| 2024-05 | AI-orchestrated multi-tool workflows | Deal screening, financial model generation | Human -> WIP | Claude tool use GA |
| 2024-06 | Cost-effective AI analysis at pipeline scale | Inbound pitch triage | WIP -> Can Do | Claude 3.5 Sonnet |
| 2024-08 | Perfect structured extraction (100% schema) | KPI benchmarking, financial model checks | WIP -> Can Do | OpenAI structured outputs |
| 2024-09 | Native PDF document processing | Pitch deck data extraction, document analysis | Human -> Can Do | Claude PDF support |
| 2024-10 | Computer-use automation for legacy tools | Technical architecture assessment | Human -> WIP | Claude computer use beta |
| 2024-10 | Multi-agent due diligence workflows | Systematic data room cross-referencing | Human -> WIP | CrewAI maturity, AutoGen |
| 2024-11 | Natural-language private company data queries | Private company data access | Human -> WIP | PitchBook Navigator + OpenAI |
| 2024-11 | Standardized AI-to-data connectivity | Portfolio company data connectivity | Human -> WIP | MCP (Model Context Protocol) |
| 2025-02 | Deep financial reasoning on demand | 3-statement model generation, DCF analysis | Human -> WIP | Extended thinking (Claude 3.7) |
| 2025-02 | Agentic coding for custom VC tools | Unit economics computation, burn rate analysis | Human -> WIP | Claude Code research preview |
| 2025-03 | Real-time market intelligence in AI | Sector trend analysis, competitive mapping | Human -> Can Do | Claude web search |
| 2025-05 | Production-grade agentic VC tooling | Report generation, exit scenario modeling | WIP -> Can Do | Claude Code GA |
| 2025-09 | End-to-end document production | Report generation (DOCX), board deck prep | Human -> Can Do | Native file generation |
| 2025-11 | AI-accessible private market data at scale | Private company data access | WIP -> Can Do | PitchBook Navigator MCP |
| 2026-02 | Full data room in single context | Systematic data room cross-referencing | Human -> Can Do | 1M token context (Claude 4.6) |
| 2026-03 | PitchBook data in conversational AI | Comparable company analysis, precedent txns | WIP -> Can Do | PitchBook + Perplexity MCP |
| 2026-03 | Fund-level return metric calculations | IRR, MOIC, DPI, TVPI, PME computation | Human -> Can Do | claude-vc v1.5.0 returns command |
| 2026-03 | XLSX export for financial outputs | Spreadsheet generation for cap tables, models | Human -> Can Do | Native file generation + skill flags |
| 2026-03 | claude-vc v1.2.0 baseline | -- | -- | Initial AI Frontier assessment |
| 2026-03 | Parallel multi-agent deal screening | Full screening with 6 concurrent agents | WIP -> Can Do | claude-vc v1.3.0 parallel agents |
| 2026-03 | Side-by-side company comparison | Structured comparison of 2-4 companies | WIP -> Can Do | claude-vc v1.3.0 vc-compare |
| 2026-03 | Customizable due diligence checklists | Stage+sector-specific DD checklist generation | Cannot -> Can Do | claude-vc v1.3.0 vc-diligence |
| 2026-03 | One-shot portfolio reporting for LPs | LP-ready portfolio summary from provided data | Cannot -> Can Do | claude-vc v1.3.0 vc-portfolio |
| 2026-04 | Single-skill consolidation (6 sub-commands) | Organizational change — no capability shift | -- | claude-vc v2.0.0 consolidation |
| 2026-04 | Self-verified financial analysis (Opus 4.7) | 3-stmt model, unit econ, DCF, consistency | WIP -> Can Do | Claude Opus 4.7 self-verification |
| 2026-04 | Opus 4.7 improved reasoning (GPQA 94.2%) | KPI auto-detection and health assessment | WIP -> Can Do | Claude Opus 4.7 extended thinking |
| 2026-04 | Opus 4.7 3.75MP vision (3x prior resolution) | High-res document and diagram analysis | -- | Claude Opus 4.7 vision upgrade |
| 2026-04 | GPT-5.4 reaches 1.05M token context window | Full data room in single context (non-Claude) | -- | GPT-5.4 (OpenAI) |
| 2026-04 | Resolve table/changelog inconsistencies | Comps, precedent txns, data room, rev proj | WIP -> Can Do | See changelog 2026-02 and 2026-03 |
| 2026-04 | Multiples valuation (simpler than DCF) | Multiples-based valuation with industry ranges | WIP -> Can Do | Python tool use, web search |
Methodology
Status definitions (expanded)
- Can Do means a domain-knowledgeable reviewer can accept or correct the output with a quick sanity check — not a substantive rewrite. For quantitative outputs (financial models, cap tables), "reliable" means the arithmetic and structure are correct given inputs. It does not mean the AI selects good assumptions — assumption quality depends on input data and human judgment.
- WIP means outputs need substantive rework, the integration is still maturing, or the capability works for simple cases but fails on edge cases or hybrid scenarios.
- Cannot (capability) means no AI tool produces useful output today.
- Cannot (regulatory) means AI could technically perform the task, but licensing, fiduciary duties, or institutional requirements prevent reliance on AI output. These items will not advance regardless of model improvements until the regulatory landscape changes. Enabler entries prefixed with "Regulatory:" mark these items in the tables.
Changelog conventions
- Dates reflect when the enabling product shipped, not when adoption matured.
- Industry entries reflect a model or product release by any provider. claude-vc entries reflect a release of this plugin specifically.
- Regressions: If a capability degrades, log it as
Can Do → WIPorWIP → Cannotwith the date and root cause. No regressions have been recorded yet — this reflects the document's youth (baseline March 2026), not an assumption that the frontier only moves forward.
Known limitations
- Enabler attribution is best-effort, not causal proof. We mean the capability became reliable around the time that enabler shipped.
- Anthropic-weighted perspective. The changelog skews toward Anthropic releases because those are what we test directly. We include other providers when their releases materially change the frontier, but coverage is not symmetric.
- Self-verification (introduced as a distinct pattern in Opus 4.7, April 2026) means the model devises checks for its own outputs — writing test assertions, re-reading generated files, cross-checking calculations — before reporting back. It catches arithmetic and structural errors but not bad assumptions.
Footnotes
-
Enabler = the AI product, feature, or integration that made (or would make) this capability possible. Enablers are either intrinsic (available to any user of the model) or integration-gated (requires a third-party subscription or MCP). Common intrinsic enablers: extended thinking, Python tool use, web search, PDF vision, native file generation, 1M context, structured output, self-verification. Common integration-gated enablers: PitchBook Navigator, Carta/Pulley, Spellbook, Harmonic, Grata, Standard Metrics/Chronograph, Hebbia Matrix. A "Can Do" item with an integration-gated enabler requires that subscription — it is not universally available. ↩
Source: claude-vc/docs/frontier.md