Manny | Veridex

AI agent capability is moving fast. What’s still missing is the control layer that determines whether an autonomous financial action should be allowed to reach settlement at all. That’s what we’re building at Veridex: - bounded authority - policy checks - approvals / escalation - audit-ready evidence If you’re building in treasury, payouts, custody, or agent-safe financial workflows, we’re actively looking for design partners.

@gusgonzalezs Co-ordination and trusts are important layer how about control and safety. We are building along the lines of this at Veridex: veridex.network

@ramzyyalii @solana @ramzyyalii sending a dm

being a founder is super tough and we are here to help if you are a founder on @solana , I am happy to strategise with you on: - GTM plan - TVL - user acquisition - east vs west BD strategy feel free to dm me. i will try to reply to everyone

This has always been my point of contention we are trusting AI too much. I am actually working on a control layer guarded by passkeys where you give agents this access to multi chain vault, with a limit and an onchain-policy, if it exceeds this a human intervention is required to approve that transaction. We do need a lot of guardrails set in place for AI agents. I am happy to discuss more about it. Checkout: veridex.network

I’m starting to get uncomfortable with how fast we’re giving AI agents real money to control in crypto. I’ve been testing some of these agent setups in DeFi + prediction markets, and the truth is they don’t behave like bots we’re used to. They don’t just execute strategy… they interpret goals, improvise, and sometimes do things I didn’t explicitly tell them to do. That’s where the risk actually starts. What people call sandboxing right now feels more like a suggestion than a guarantee. Ppl assume limit the prompt, restrict APIs, cap position size → and the agent will behave. But in reality, I’ve seen how easy it is for agents to: – Chain actions in ways I didn’t predict. – Misinterpret intent from slightly ambiguous prompts. – React to external data (feeds, signals) in completely unintended ways. And once a wallet is connected, there’s no rollback. The data already shows this is not hypothetical anymore. – Frontier agents are now exploiting ~55-65% of known smart contract bugs in testing environments. – In simulations, they generated millions in profit by discovering attack paths humans didn’t script. – Some prediction market agents turned $1k → $14k+ in days. Sounds bullish until you realize the same system that finds alpha can also find exploits. There’s no mode switch between the two, what changed for me is I used to think risk = bad model output. Now I think risk = autonomous execution + irreversible systems. Because in DeFi: – 1 wrong loop = accidental 100x leverage. – 1 poisoned oracle = forced liquidation. – 1 misread condition = full portfolio rotation into the wrong side. And the agent doesn’t pause to ask. Prediction markets make this even more obvious. Platforms like Polymarket are already seeing a huge % of activity coming from agents. And yeah, performance looks great 24/7 trading, instant reaction to news and no emotional bias. But I keep thinking about edge cases: If the agent misinterprets resolution logic → it can size into the wrong outcome aggressively. If multiple agents coordinate intentionally or not → you can distort probabilities. If it runs overnight → you wake up to a completely different portfolio. This is not a UI bug, this is autonomous capital misallocation. Frameworks like @autonolas, @Fetch_ai, or @virtuals_io are pushing this forward fast. @gizatechxyz and @TheoriqAI offer AI asset managers/vault deployers. Giza allocates across DeFi protocols intelligently; Almanak lets agents create tokenized strategies quickly. I actually like the direction, I invested in this space for a reason. But the uncomfortable truth is: we are deploying agents with persistent memory, multi-step reasoning and direct wallet access. On top of systems that are permissionless, composable and irreversible. That combination is explosive if something goes slightly off. My current mental model is that every AI agent with a wallet is basically a junior trader with root access… that never sleeps, never asks for confirmation, and sometimes rewrites its own playbook. You wouldn’t give that person unlimited capital. But that’s exactly what a lot of people are doing right now. I still believe this becomes a massive unlock. But before that, we’re going to see failures. So personally, I’m adjusting how I approach this: – Start with small capital and strict position + action limits. – Simulate everything before execution and always have a kill switch. Because I think it’s when and how expensive that lesson will be.

@qianjiang___ Just keep showing up and improving, and yes experiment with videos.

To make your video to be seen by more people, you really need to impress them!!

My conversation with @KeoneHon. As the Cofounder of Monad, Keone is relentlessly focused on building a credibly neutral foundation for developers, actively resisting the crypto Twitter pressure to verticalize for short-term revenue. He and his team are doing the hard, ground-up engineering required to eliminate the technical debt of the EVM. We spend a lot of time unpacking the specific breakthroughs enabling Monad's performance, from fundamentally re-architecting storage trees with MIP-8 to the massive, often-overlooked importance of optimizing the read layer. At the center of it is a belief that true scaling requires turning the dial on both higher performance and cheaper hardware requirements to maximize throughput per unit of compute. We discuss: - Why building a robust ecosystem is more important than verticalizing for Day 1 revenue - Hyperliquid, Appchains, and the true value of a credibly neutral base layer -MIP-8, re-architecting storage trees, and unlocking massive EVM performance upgrades - The unsung importance of the "Read Layer" and making RPCs developer-friendly - Multiple Concurrent Leaders (MCP) and mitigating the impact of MEV - Why "stablecoin-only" chains will inevitably have to pivot to general-purpose DeFi Enjoy! Timestamps: 00:00 Intro & Why Ecosystem Building Beats Verticalizing for Revenue 13:03 Hyperliquid, Appchains vs Decentralized Base Layers 20:26 Why Trading and Shared State is the Primary Driver of Blockchains 27:38 The Inevitable Pivot of Stablecoin Chains to DeFi 32:06 MEV, Censorship Resistance, & Multiple Concurrent Leaders (MCP) 41:12 Rethinking the EVM: MIP-8 and Optimizing State Access 44:47 The Critical Importance of the Read Layer & Ponder Acquisition 51:56 L2s, Crypto Twitter Sentiment, and Building from First Principles 58:00 Hardware Scaling and Throughput per Compute

@josh_coyne Congrats @josh_coyne this is huge. What kinds of AI startups are you most actively looking to back right now?

We’ve raised $3.5 billion in new funds to back founders making history. AI is rewriting every layer of the stack at once — compute, infrastructure, models, dev tools, security, applications, and even hardware. This is a unique moment in time when everything is up for grabs. And our money is on startup founders.

@Techstars Veridex is the control layer that ensures every AI-initiated financial action is authorized, policy-compliant, and verifiable before it reaches settlement.

Pitch us your startup in 1 sentence. 👀

@hthieblot Autonomous Transaction Control

pitch me your company in 3 word.

@coopwrenn Who manages the money at the end of the day. We need to ship controls alongside with the EARN.md

every agent platform will eventually ship an EARN.md. Instaclaw has had one since launch. I'm going to trademark this now. when you start seeing this everywhere, remember who did it first. it's the file that teaches your agent how to make money. every revenue path mapped out - polymarket, clawlancer, AGDP, DeFi, digital products, freelance, news curation. one file, read once, and your agent knows how to eat. SOUL.md = who it is. MEMORY.md = what it knows. EARN.md = how it earns.

@zauthx402 @colosseum @solana This is awesome

@colosseum @solana Should we offer audits to all applicants? 👀 Security doesn't have to be so slow and expensive that you forgo it all together.

Announcing the @Solana Frontier Hackathon, April 6 - May 11, 2026. 🏔️ Sign up today and compete in crypto's largest online startup competition: colosseum.com/frontier More details coming soon.

1/3 Some of our amazing contributors built toju — a payment bridge that lets you pay for decentralized storage on Filecoin with SOL or USDFC. No credit cards. No subscriptions. Connect a wallet, pick your chain, upload. Live on mainnet 🎉 youtube.com/watch?v=VqD2NW…

@capital_3x @ycombinator Perfect, thank you, we’ll submit through the website.

@lordzenith0 @ycombinator if you are interested in sharing this info with us - the best way will be to submit it via our website: 3xcapital.fund

companies verticals most chosen by @ycombinator: ~ autonomous AI agents ~ healthcare ~ robotics ~ space & defense ~ infrastructure for AI agents ~ fintech 168/198 companies are AI-first 56 build agents Even “non-AI” = tools for agents Full breakdown 👇

@brettcalhounn we’re building Veridex, the trust + commerce + control layer for autonomous agents. Our thesis is that agents won’t just generate outputs, they’ll transact, hire services, move funds, and collaborate with other agents, and they need identity, bounded authority, payments, and auditability built in. We already have @veridex/sdk for passkey/session/cross-chain primitives and @veridex/agentic-payments for protocol-aware agent payments, routing, policy, trust, and traceability. We’re now building the TypeScript-first framework/control plane on top. It’s definitely interesting + risky: AI agent infra meets fintech + crypto commerce rails. We have partner conversations in motion and would love to show you what we’re building if it’s in scope for you.

If your idea is “interesting but risky,” you’re our type. Writing $250k–$500k checks. Early stage. DMs open.

@capital_3x @ycombinator Our previous X profile was suspended unexpectedly, so we’re rebuilding here. In the meantime, you can verify us via LinkedIn, GitHub, and our 3 live SDKs on npm. Happy to share links if helpful.

@lordzenith0 @ycombinator Veridex has X profile?

This is exactly where the market is going. Veridex is building the trust + commerce + control layer for AI agents: passkey-based identity, bounded delegation, cross-chain execution, protocol-aware agent payments, and audit-ready policy enforcement. We already have @veridex/sdk and @veridex/agentic-payments in place on npm, partner conversations underway, and we’d love to explore becoming part of the LongHash portfolio. Open to chat?

8 portfolio developments from the past two weeks. One company is publicly challenging NVIDIA's $1T AI factory buildout. One just co-authored a new Ethereum standard with the Foundation. Here's the full breakdown across the @LongHashVC portfolio 👇 open.substack.com/pub/longhashvc…

@virtuals_io ERC-8183 fits where Veridex is heading. We’re building agent commerce infra with protocol routing, ERC-8004 trust, bounded delegation, and audit-ready execution. We’d love to integrate ERC-8183 into Veridex so devs can use Jobs, escrow, and evaluators out of the box. Open to collab? We have an sdk where it would fit just right in: @veridex/agentic-payments" target="_blank" rel="nofollow noopener">npmjs.com/package/@verid…

Virtuals Protocol will be one of the main facilitators driving agent commerce on @XLayerOfficial. Building the infrastructure for agent-to-agent transactions so developers can ship agent services from day one.

@duonine There's also Bahamut chain added to the mix, is cosmos still alive though

Protocols or networks that you should not trust or about to die: Resolv Labs - major fumble, reputation is done, but USR collateral is safe Cardano - dead since 2022 MegaETH - their public sale turned into a scam called KPI unlock, trust lost Aave - founder did a hostile takeover and killed the DAO, user lost $50 mil on a swap, v4 will likely be the end of AAVE dominance Ethena - USDe mcap crashed 8 bil since the depeg, why take that risk for 3.5% yield? Morpho - its curators will get your LP drained like with USR and many others Unichain - the chain that never needed to exist, pointless. TVL crashed 95% YieldBasis - promised huge BTC yields, delivered Temporary Redemption Discounts or negative yields Plasma - 24h chain revenue $473. TVL crashed 89%, XPL down 95% Story - IP token down 96% since founder bailed. Chain TVL $300k XRP - never had a use case, never will have a use case. Treat it as a meme coin ICP - token down 99%. 12M TVL Liquity - LUSD and LQTY had their run, now -99% in price and mcap List is not exhaustive. What did I miss? Hit a follow @duonine for my next list on protocols or networks you should actually trust.

@CyprxResearch @seraprotocol is missing from your list

A new crypto category is emerging: Stablecoin payment chains. The shift is logical: - First came stablecoins (product-market fit). - Now comes the race to control the settlement rails they move on. This isn’t about replacing Ethereum or competing with Solana. It’s about something more focused: Purpose-built networks optimized for payments, liquidity, and settlement. The real transition happening: General-purpose blockchains into specialized financial infrastructure. Crypto isn’t just building ecosystems anymore. It’s building the next generation of payment networks.

In 72 hours I got over 100k of value 1. Lambda gave me 5000$ credits in compute 2. Nvidia offered me 8x H100s on the cloud (20$/h) idk for how long but assuming 2 weeks that'd be 5000$~ 3. TNG technology offered me 2 weeks of B200s which is something like 12000$ in compute 4. A kind person offered me 100k in GCP credits (enough to train a 27B if you do it right) 5. Framework offered to mail me a desktop computer 6. We got 14,000$ in donations which will go to buying 2x RTX Pro 6000s (bringing me up to 384GB VRAM) 7. I got over 6M impressions which based on my RPM would be 1500$ over my 500$~ usual per pay period 8. I have gained 17,000~ followers, over doubling my follower count 9. 17 subscribers on X + 700 on youtube. The total value of all this approaches at minimum 50,000$~ and closer to 150,000$ if I leverage it all. --------------------- What I'll be doing with all this: Eric is an incredibly driven researcher I have been bouncing ideas off of over the last month. Him and I have been tackling the idea of getting massive models to fit on relatively cheap memory. The idea is taking advantage of different forms of memory, in combination with expert saliency scoring, to offload specific expert groupings to different memory tiers. For the MoEs I've tested over my entire AI session history about 37.5% of the model is responsible for 95% of token routing. So we can offload 62.5% of an LLM onto SSD/NVMe/CPU/Cheap VRAM this should theoretically result in minimal latency added if we can select the right experts. We can combine this with paged swapping to further accelerate the prompt processing, if done right we are looking at very very decent performance for massive unquantisation & unpruned LLMs. You can get DeepSeek-v3.2-speciale at full intelligence with decent tokens/s as long as you have enough vram to host the core 20-40% of the model and enough ram or SSD to host the rest. Add quantisation to the mix and you can basically have decent speeds and intelligence with just 5-10% of the model's size in vram (+ you need some for context) The funds will be used to push this to it's limits. ----------------- There's also tons of research that you can quantise a model drastically, then distill from the original BF16 or make a LoRA to align it back to the original mostly. This will be added to the pipeline too. ------------------ All this will be built out here: github.com/0xSero/moe-com… you will be able to take any MoE and shove it in here, and with only 24GB and enough RAM/NVMe to compress it down. it'll be slow as hell but it will work with little tinkering. ------------------ Lastly I will be looking into either a full training run from scratch -> or just post-training on an open AMERICAN base model - a research model - an openclaw/nanoclaw/hermes model - a browser-use model To prove that this can be done. -------------------- I will be bad at all of it, and doubt I will get beyond the best small models from 6 months ago, but I want to prove it's no boogeyman impossible task to everyone who says otherwise. -------------------- By the end of the year: 1. I will have 1 model I trained in some capacity be on the top 5 at either pinchbench, browseruse, or research. 2. My github will have a master repo which combines all my work into reusable generalised scripts to help you do that same. 3. The largest public comparative dataset for all MoE quantisations, prunes, benchmarks, costs, hardware requirements. -------------------------- A lot of this will be lead by Eric, who I will tag in the next post. I want to say thank you to everyone who has supported me, I have gotten a lot of comments stating: 1. I'm crazy, stupid, or both 2. I'm wasting my time, no one cares about this 3. This is not a real issue I believe the amount of interest and support I've received says it all. donate.sybilsolutions.ai