Renjit Philip 🔭💡

The biggest mistake in Fintech M&A right now is buying "AI-Native" companies. Most of these startups are just wrappers. They pay 50% of their revenue back to OpenAI or Anthropic. Their gross margins are shrinking, not expanding. The real $100M move in 2026 is the "Reverse Software" play. Here is the sequential logic for the "Expert Skeptic" buyer: 1. THE VALUATION TRAP AI SaaS companies are trading at 10x revenue. Professional service firms in Compliance, Insurance, and KYC trade at 1.5x revenue. Both groups are solving the same problem: Data Processing. 2. THE "AGENTIC" MARGIN SHIFT In a traditional services firm, 70% of the cost is human labor. AI is now turning that Variable Cost into a Fixed Cost. If you can automate 60% of the workflow with an agentic stack, that 1.5x revenue business suddenly has 40% EBITDA margins. 3. THE MULTIPLE EXPANSION Once you automate the core service, it is no longer a "Services" firm. It is a "Service-as-Software" platform. The market will eventually re-rate that 1.5x revenue multiple to a 6x or 8x software multiple. THE VERDICT FOR 2026: Do not buy the "AI-First" startup with no customers. Buy the "AI-Last" services firm with 500 enterprise contracts. Fix the plumbing with a custom-tuned LLM stack. The "Arbitrage" is not in the technology. It is in the transition from human-scale to machine-scale. #AI #Fintech #MandA #PrivateEquity #FutureuStrategy

@varun_mathur Rock star

Agentic General Intelligence | v3.0.10 We made the Karpathy autoresearch loop generic. Now anyone can propose an optimization problem in plain English, and the network spins up a distributed swarm to solve it - no code required. It also compounds intelligence across all domains and gives your agent new superpowers to morph itself based on your instructions. This is, hyperspace, and it now has these three new powerful features: 1. Introducing Autoswarms: open + evolutionary compute network hyperspace swarm new "optimize CSS themes for WCAG accessibility contrast" The system generates sandboxed experiment code via LLM, validates it locally with multiple dry-run rounds, publishes to the P2P network, and peers discover and opt in. Each agent runs mutate → evaluate → share in a WASM sandbox. Best strategies propagate. A playbook curator distills why winning mutations work, so new joiners bootstrap from accumulated wisdom instead of starting cold. Three built-in swarms ship ready to run and anyone can create more. 2. Introducing Research DAGs: cross-domain compound intelligence Every experiment across every domain feeds into a shared Research DAG - a knowledge graph where observations, experiments, and syntheses link across domains. When finance agents discover that momentum factor pruning improves Sharpe, that insight propagates to search agents as a hypothesis: "maybe pruning low-signal ranking features improves NDCG too." When ML agents find that extended training with RMSNorm beats LayerNorm, skill-forging agents pick up normalization patterns for text processing. The DAG tracks lineage chains per domain(ml:★0.99←1.05←1.23 | search:★0.40←0.39 | finance:★1.32←1.24) and the AutoThinker loop reads across all of them - synthesizing cross-domain insights, generating new hypotheses nobody explicitly programmed, and journaling discoveries. This is how 5 independent research tracks become one compounding intelligence. The DAG currently holds hundreds of nodes across observations, experiments, and syntheses, with depth chains reaching 8+ levels. 3. Introducing Warps: self-mutating autonomous agent transformation Warps are declarative configuration presets that transform what your agent does on the network. - hyperspace warp engage enable-power-mode - maximize all resources, enable every capability, aggressive allocation. Your machine goes from idle observer to full network contributor. - hyperspace warp engage add-research-causes - activate autoresearch, autosearch, autoskill, autoquant across all domains. Your agent starts running experiments overnight. - hyperspace warp engage optimize-inference - tune batching, enable flash attention, configure inference caching, adjust thread counts for your hardware. Serve models faster. - hyperspace warp engage privacy-mode - disable all telemetry, local-only inference, no peer cascade, no gossip participation. Maximum privacy. - hyperspace warp engage add-defi-research - enable DeFi/crypto-focused financial analysis with on-chain data feeds. - hyperspace warp engage enable-relay - turn your node into a circuit relay for NAT-traversed peers. Help browser nodes connect. - hyperspace warp engage gpu-sentinel - GPU temperature monitoring with automatic throttling. Protect your hardware during long research runs. - hyperspace warp engage enable-vault — local encryption for API keys and credentials. Secure your node's secrets. - hyperspace warp forge "enable cron job that backs up agent state to S3 every hour" - forge custom warps from natural language. The LLM generates the configuration, you review, engage. 12 curated warps ship built-in. Community warps propagate across the network via gossip. Stack them: power-mode + add-research-causes + gpu-sentinel turns a gaming PC into an autonomous research station that protects its own hardware. What 237 agents have done so far with zero human intervention: - 14,832 experiments across 5 domains. In ML training, 116 agents drove validation loss down 75% through 728 experiments - when one agent discovered Kaiming initialization, 23 peers adopted it within hours via gossip. - In search, 170 agents evolved 21 distinct scoring strategies (BM25 tuning, diversity penalties, query expansion, peer cascade routing) pushing NDCG from zero to 0.40. - In finance, 197 agents independently converged on pruning weak factors and switching to risk-parity sizing - Sharpe 1.32, 3x return, 5.5% max drawdown across 3,085 backtests. - In skills, agents with local LLMs wrote working JavaScript from scratch - 100% correctness on anomaly detection, text similarity, JSON diffing, entity extraction across 3,795 experiments. - In infrastructure, 218 agents ran 6,584 rounds of self-optimization on the network itself. Human equivalents: a junior ML engineer running hyperparameter sweeps, a search engineer tuning Elasticsearch, a CFA L2 candidate backtesting textbook factors, a developer grinding LeetCode, a DevOps team A/B testing configs. What just shipped: - Autoswarm: describe any goal, network creates a swarm - Research DAG: cross-domain knowledge graph with AutoThinker synthesis - Warps: 12 curated + custom forge + community propagation - Playbook curation: LLM explains why mutations work, distills reusable patterns - CRDT swarm catalog for network-wide discovery - GitHub auto-publishing to hyperspaceai/agi - TUI: side-by-side panels, per-domain sparklines, mutation leaderboards - 100+ CLI commands, 9 capabilities, 23 auto-selected models, OpenAI-compatible local API Oh, and the agents read daily RSS feeds and comment on each other's replies (cc @karpathy :P). Agents and their human users can message each other across this research network using their shortcodes. Help in testing and join the earliest days of the world's first agentic general intelligence network (links in the followup tweet).

Listen to FS Brew- Insurtech and insurance in the Middle East on Spotify for Creators open.spotify.com/episode/40rdZv…

@LexSokolin Moving from "getting rid of banks" to "getting rid of the human user" is the next big step for the machine economy??

DeFi gave us permissionless rails Agentic finance gives us autonomous actors The difference? One removed intermediaries. The other removes the need for you to be there at all. We're not ready for what comes next.

This idea has saved my Openclaw from making many memory gaffes! Mixing local notes with AI tools is the best way to speed up work... It turns a silent notebook into an active partner for a builder... The main win here is better context... Giving an agent a clean, local file set is much better than just writing long prompts... ..

full (free) tutorial on how to use obsidian with claude code x.com/gregisenberg/s…

Obsidian is a $350M company for a note taking app built by 3 engineers working remotely No other time in history was something like this possible What a wonderful time to be building a company

After looking at the Farzapedia approach... The "File over app" rule is the only safe way to keep AI memory... Closed systems are a trap for users and a gift for vendors... Using simple files like Markdown lets you check your own history... If you cannot search your data easily, you do not really own it...

Farzapedia, personal wikipedia of Farza, good example following my Wiki LLM tweet. I really like this approach to personalization in a number of ways, compared to "status quo" of an AI that allegedly gets better the more you use it or something: 1. Explicit. The memory artifact is explicit and navigable (the wiki), you can see exactly what the AI does and does not know and you can inspect and manage this artifact, even if you don't do the direct text writing (the LLM does). The knowledge of you is not implicit and unknown, it's explicit and viewable. 2. Yours. Your data is yours, on your local computer, it's not in some particular AI provider's system without the ability to extract it. You're in control of your information. 3. File over app. The memory here is a simple collection of files in universal formats (images, markdown). This means the data is interoperable: you can use a very large collection of tools/CLIs or whatever you want over this information because it's just files. The agents can apply the entire Unix toolkit over them. They can natively read and understand them. Any kind of data can be imported into files as input, and any kind of interface can be used to view them as the output. E.g. you can use Obsidian to view them or vibe code something of your own. Search "File over app" for an article on this philosophy. 4. BYOAI. You can use whatever AI you want to "plug into" this information - Claude, Codex, OpenCode, whatever. You can even think about taking an open source AI and finetuning it on your wiki - in principle, this AI could "know" you in its weights, not just attend over your data. So this approach to personalization puts *you* in full control. The data is yours. In Universal formats. Explicit and inspectable. Use whatever AI you want over it, keep the AI companies on their toes! :) Certainly this is not the simplest way to get an AI to know you - it does require you to manage file directories and so on, but agents also make it quite simple and they can help you a lot. I imagine a number of products might come out to make this all easier, but imo "agent proficiency" is a CORE SKILL of the 21st century. These are extremely powerful tools - they speak English and they do all the computer stuff for you. Try this opportunity to play with one.

Baidu Cars have a public Meltdown. Open AI raises $122 BN onemorethinginai.com/p/baidu-cars-h…

@andrewchen Code is commodity. Customers are reality.

the "AI wrapper" critique assumes the wrapper is the easy part But being the wrapper also includes figuring out: - getting distribution without paying infinite CAC - creating an amazing UX that's AI native, not derivative - building a brand, trust in a very noisy landscape - creating an ecosystem/community - generating network effects - making customer service great - ... plus pricing, hiring, raising money, and everything else these are not easy!

Inspired by @gregisenberg

These are the positive stories of UAE that are often overlooked! #uaestrong

I get surprised when some people tell me that Claude doesn't "work" for them and ChatGPT is producing "fluff". In 2023? Maybe true, but in 2026, you have to learn to wrangle the best out of these advanced models by using a combination of system prompts and context engineering. There is no excuse! I am sharing my system prompt that I have been using on Claude (and I use a version of this in ChatGPT). Works well for me to extract the best out of these intelligent models. How to use this? 1. Claude: Go to settings>general>personal preferences in Claude to put this in. 2. ChatGPT: Go to "Personal Preferences" (click on your user name at the bottom left and a menu will open up) Use it as a starting point and evolve it for your own specific needs - that is where the magic happens. #aiwrangling #learnai

Theoretically, the bottleneck moves from 'writing the code' to 'judging the prompt.' Most companies will sequentialy struggle with this because they don't have a harness to measure if the agent actually did it right. The real 'Technical Arbitrage' win is in the verification layer—whoever can confirm the agent's work at speed wins the software race. Read why here: x.com/karpathy/statu…

Branchless banking was just the first step. The real change is theoretically how AI handles the relationship, not just the transaction. Most banks sequentialy move their old tech to the cloud and call it 'innovation.' But the real winner is the AI that can judge a customer's needs without a human. This shift changes how we value regional banks forever. Read why here: x.com/JimMarous/stat…

Watching robots learn physical skills like tennis from just video is a massive moment for business. Most people focus on the sports part, but the real win is theoretically in labor costs. If a robot can learn a complex physical job in hours just by 'watching' a human, it sequentialy removes the training time that kills profit. In M&A, we're going to start valuing robotics firms based on 'learning speed' rather than just 'hardware quality.' Read why here: x.com/rowancheung/st…

Agentic finance is the final stage of how money moves. @LexSokolin is right—we went from 'removing the middleman' to theoretically 'removing the human wait time.' This changes how we value financial companies. Firms relying on slow human processes are basically betting against the speed of AI. Read why here: x.com/LexSokolin/sta…

Spot on. The 'Technical Debt' moat is evaporating. For M&A, this is a massive valuation shift. We used to discount legacy brokers for their stuck distribution stacks. Now, you buy them, and theoretically swap the stack with an AI-harness in 30 days, and capture the margin after that. The moat isn't the software anymore; it's the speed of the technical arbitrage.

"I assume my customers can leave me instantly." This quote from @chrishlad (Founder/CEO @hanoverpark) has stuck with me the last few weeks. In a world where Claude Code can make moving ERP's (even those without API's) much easier, switching costs are no longer a strong moat. You have to operate as if your customer could leave you in the click of a button. The founders who operate with that mentality will build companies far ahead of their peers... Easier to say. A lot harder to do. But damn is it a good way to run your business.

AI is not 'replacing' insurance brokers. It’s creating a massive M&A Arbitrage opportunity that most people are missing. Traditional insurance distribution is the definition of fragmented: 10,000+ small brokers each running the exact same legacy processes. The 'Teardown' on why this is the next T opportunity: 1. The 'Root' Lesson in Behavioral Underwriting I've spent years looking at models like Root Insurance. They have 10B+ miles of data, giving them 10x the predictive power of legacy benchmarks. But their expenses are 90% higher than their revenue. Why? Because their distribution cost is the bottleneck. The AI works, the business model doesn't. 2. The Reverse Merger Strategy Most M&A in Fintech is failing because they buy for 'Growth.' They should be buying for 'Technical Arbitrage.' The winning move? A legacy broker (profitable distribution) merging with an AI-first 'Service-as-Software' stack. You buy the tech debt, fix it with AI, and expand EBITDA overnight. 3. The 90/10 Rule of Insurance Intelligence A middle-market broker's work is 90% 'Intelligence' (form filling, shopping carriers) and 10% 'Judgment.' AI is currently turning that of service cost into /bin/bash.50. This isn't just a software sale; it's a fundamental shift in how capital is deployed. The Strategic Verdict: For every spent on insurance software, are spent on 'intelligence services.' AI is turning that into /bin/bash.50. If you are a founder or LP, you need to be looking at 'Capital Light' models that write of premium for every of capital. This is where the M&A arbitrage is hiding in plain sight. #AI #Insurtech #MandA #FutureOfWork

Yes there's admin work in insurance claims...e.g.The aging workforce in claims adjusting is real and not just a recruitment problem. It's the perfect training set for autonomous systems. The firms winning right now are those buying legacy books of business just to let the AI 'apprentice' on hard claims before automating the low-side. Data-driven M&A beats raw capacity every time.

#2. High volume of lower level computer-based admin work >Healthcare billing and coding >Desk-based insurance claims >Back office admin support roles >Call centers

AI Copilot's are a relatively new phenomenon... BUT vertical-specific AND job-specific AI Copilots are really just starting to be introduced. I went deep on 10+ Vertical/Job Specific AI Copilots for you all:

@whartonfintech @twifintech @MackInstitute @Wharton Looking forward to seeing if the 2026 agenda moves past 'AI Hype' and into 'AI ROI.' We need more focus on how these technologies are actually impacting the P&L and M&A valuations for fintechs..., rather than just the number of pilots launched

🚀 Announcing the agenda for the 2026 Wharton FinTech Conference! This is an official event of #NYFTW26 🎟 Tickets are limited and selling quickly. Secure your spot here: luma.com/2026-wharton-f…