Intelligence dispatches from a 88,359-entity knowledge graph, 321 startup postmortems, and a 12-persona Analytical Council. Published 5 days a week. Free.
For shorter daily perspectives from LinkedIn, see Insights.
Harvey raised $200M at an $11B valuation in March 2026 — then launched a free training academy for lawyers. That sequence is not a coincidence. The enterprise AI moat is no longer the model. It's the trained user base.
Read note →Humane raised $230M, achieved a $700M+ peak valuation, and sold to HP for $116M in February 2025. The postmortem reveals a death pattern that is repeating across AI hardware startups — and the structural mistakes that founders and investors must stop making.
Read note →India's AI sector drew ₹2,110 crore in Q1 2026 — a 73% YoY surge that moved it to #3 in overall startup funding. But the headline hides a more interesting structural story about where the capital is actually going.
Read note →Midjourney generates $4.7M in revenue per employee. Cursor crossed $2B ARR with fewer than 150 people. AI-native revenue density is redrawing the valuation map — and most investors are still using the wrong ruler.
Read note →OpenAI launched Operator and Atlas, Perplexity gave Comet away free, Atlassian paid $610M for The Browser Company, and Google embedded Gemini in Chrome — all in fifteen months. The convergence tells you which interface layer the next decade is being fought over.
Read note →When agents handle 95% of analytical work, human judgment atrophy becomes a systemic risk. The solution isn't less AI — it's a deliberate exercise programme for human decision-makers.
Read note →Builder.ai raised $450M, hit a $1.5B valuation, and collapsed in May 2025. The 'AI' was 700 humans. The $220M of 2024 revenue was actually $55M. Here are the four signals that should have caught it in 2019.
Read note →Harvey at $11B, Legora at $5.6B, Microsoft shipping a Word Legal Agent, an open-source clone hitting Hacker News — the legal AI market structure was redrawn in eight days at the end of April 2026. Here is what the new map looks like.
Read note →Cursor at $2B ARR with ~150 people. Lovable at $500M ARR with 146. Midjourney at $5M-per-employee. The AI-native cohort just shattered software's 30-year revenue density ceiling.
Read note →Three companies in three regulated verticals shipped multi-tier agent architectures within twelve months. They didn't share code. They converged on the same shape because the problem only has one solution.
Read note →Only 14% of enterprise AI agent pilots reach production scale. The 37-point gap between lab benchmark and real-world performance is not a model problem — it's an evaluation infrastructure problem.
Read note →Jasper went from $120M ARR to $35M in eighteen months. Frontier API prices fell 60–80%. The 2023 wrapper cohort is dying for the same reason every commoditised reseller dies — declining marginal margin meets a free competitor.
Read note →Two startups funded a combined $24M are turning the $500K commercial due diligence engagement into a $50K, 24-hour deliverable. The analysis layer of management consulting is being repriced in real time.
Read note →India's 958 million internet users live in 600,000 villages. The companies winning at scale aren't going direct — they are equipping the intermediaries who already have trust.
Read note →Gartner predicts 40% of agentic AI projects will be canceled by 2027. Only ~130 of thousands of 'AI agent' vendors are building genuinely agentic systems. The gap between hype and production is the signal — and it points to where the real money flows next.
Read note →78% of enterprises have AI agent pilots running. 14% reach production. Stanford's Enterprise AI Playbook traces 95% of failures to organisational readiness, not technology. The pilot-to-production gap is an operations problem masquerading as a technology problem.
Read note →CB Insights data shows 43% of VC-backed startup deaths trace to product-market fit failure. The TAM ceiling variant — where the product works but the addressable market can't support the valuation — killed Quibi ($1.75B raised, shut down in 6 months) and Zilingo ($300M raised, liquidated 2023).
Read note →Activate Fund's $75M debut fund targets early-stage Indian AI startups. With 958M internet users and GDP at $4.15T, the application layer remains wide open — and most Indian VCs are still funding infrastructure.
Read note →An estimated $800B in SaaS market cap eroded in 2024–2025. The reset isn't temporary — it's structural. Three tiers are crystallising. Only one has venture economics.
Read note →AI agents should handle analysis. Humans should handle decisions. The distinction sounds obvious until you see how many organisations get it backwards.
Read note →Most AI teams are shipping blind. Karpathy's autoresearch made the pattern explicit: define a metric, measure before touching anything, reject changes that lower the score. This is how AI products compound.
Read note →The operational ceiling of any company isn't capital or talent — it's humans in the process. Zero-human deployment removes that ceiling entirely. Here's what it looks like in practice.
Read note →Gartner projects 40% of agentic AI projects cancelled by 2027. Only 12% of enterprises expect ROI from agents within three years. The failures aren't in the models — they're in the seams between agents.
Read note →Manthan Intelligence's knowledge graph contains 220 startup postmortems. AI-first companies die in predictable ways. Here are the 10 patterns, with frequencies.
Read note →Only 14% of India's MSMEs access formal credit. Women entrepreneurs face a structural pricing failure — and the data shows they're systematically lower risk than the market charges them for. Here's where the alpha is.
Read note →Meesho built India's 4th most valuable startup on zero seller commissions. What that inversion reveals about margin logic in emerging markets — and what it means for the next 500 marketplace founders.
Read note →Nathan Benaich built Europe's largest solo GP fund with AI drafting tools and no analyst team. His stack reveals the boundary between productivity multiplier and institutional intelligence.
Read note →The foundation model layer is a commodity. The real value migration in AI is moving to whoever builds institutional memory first — and knowledge graphs are how you get there.
Read note →We ran a blind assessment on Graphcore using only December 2020 data. Then we checked what actually happened. Here's the full record.
Read note →Agentic AI will grow from $5.2B to $197B at 43.8% CAGR. The entire software industry is misunderstanding the pricing model. Copilot pricing will not capture agent value. The winner is whoever figures out labour replacement pricing first.
Read note →Nature Medicine published a landmark study: ChatGPT Health under-triaged 52% of gold-standard emergencies. Single-agent systems fail catastrophically in high-stakes decisions.
Read note →The full argument for why knowledge work needs structured disagreement, compounding memory, and autonomous calibration — not another chatbot wrapper. By Mayank Mathur.
Read note →Three signals converged this week. India's fintech sector is entering IPO territory. Private credit is on course for its biggest year in emerging markets. Angel investment deal volume declined 44% year-over-year. The capital stack is reshaping.
Read note →Single-agent AI systems fail at critical decisions — healthcare proved it, finance is next. Multi-agent architecture with structured deliberation is the future of venture capital.
Read note →Of 196 startup postmortems, 13 are fintech. Five repeatable patterns explain why India's most funded sector has the worst allocation efficiency.
Read note →India agritech will grow from $9B to $28B by 2030, yet no VC-backed startup has IPO'd. What the $2.9B funding trail reveals about exit risk and opportunity.
Read note →Competition kills 34% of startups — not product-market fit, not unit economics. We analysed 188 postmortems across 13,499 companies to find the mechanical causes of startup death.
Read note →We spent a year building an AI that analyses startups the way a team of 12 specialist analysts would — except it never forgets, never gets tired, and cross-references every company against 13,499 others. Now we're publishing what it finds.
Read note →Free dispatches, Mon-Fri. Unsubscribe anytime.
No spam. Just intelligence.