Do you remember when you joined X? I do! #MyXAnniversary

AI coding agents can build software quickly. But they can also say: “Done.” “Deployed.” “Everything is working.” …while the database is still disconnected, an endpoint is failing, or the contract was never properly deployed. I kept running into this while building with AI agents. So for the Spark hackathon, I built ShipReceipt. ShipReceipt checks the actual project: → public GitHub repository → selected commit → live deployment → health and readiness endpoints → deployed Monad contract It then records exactly what passed and what failed. No AI decides the result. No fake green success screen. No hiding failed checks. One of the live receipts passed 4 out of 5 checks because the readiness endpoint returned a real HTTP 404. ShipReceipt marked it Partial and preserved the failure in the evidence. The detailed evidence stays offchain, while its integrity, result, issuer, and timestamp are recorded on Monad thus making later changes detectable. This can help developers verify AI-generated work, clients inspect software before accepting delivery, and hackathon judges review working submissions more confidently. AI can claim it shipped. ShipReceipt proves what actually worked. ShipReceipt is now live on the Build Anything showcase. Check out the project, test the live receipts, and please upvote it if you believe software completion should come with verifiable evidence: buildanything.so/showcase/sh… Live app: shipreceipt.vercel.app Partial receipt: shipreceipt.vercel.app/recei… Built on Monad Testnet. #Monad #BuildOnMonad #SparkHackathon #BuildInPublic #AIAgents

I use AI coding agents almost every day. They help me move faster, but I’ve noticed a recurring problem: The agent says the work is complete, but I still have to manually verify everything. Did the deployment actually work? Is the database connected? Does the health endpoint return 200? Is the contract really deployed? Did the transaction confirm? That experience led me to build ShipReceipt for the Spark hackathon. ShipReceipt verifies a live software build, binds the result to a specific GitHub commit, creates a tamper-evident evidence record, and anchors the receipt on Monad. It does not reward confident claims. It checks the actual system. So when a build is only partly working, ShipReceipt says Partial instead of presenting a fake green success screen. I believe this becomes increasingly useful as more software is built by coordinating AI agents. We are producing code faster. Now we also need better ways to prove what was actually shipped. AI can claim it shipped. ShipReceipt proves it. #Monad #BuildOnMonad #SparkHackathon #BuildInPublic #AIAgents #SoftwareEngineering

So my mutuals can now see me?

Part 3: Campaigns vs Workflows vs Transactional Emails. This is the simple framework that makes Loops easier to understand: Campaigns = broadcasts Workflows = user journeys Transactional = product-triggered emails Once you see the difference, the product makes more sense.

Part 2: Why SaaS products need more than a newsletter tool. The main idea: A SaaS product does not only send updates. It sends onboarding emails, product nudges, password resets, receipts, alerts, reactivation emails, and lifecycle messages. That is a different email problem.

Applying for the Product Educator role at @loops. I made a 3-part video series showing how I’d explain Loops to SaaS founders, developers, and product teams. The goal was not to make an ad. The goal was to make the product category clearer: modern SaaS email is not just newsletters. Part 1: What is Loops?

Can't eeven parse this!

I've become increasingly interested in content production as a system. Not just editing videos. But designing repeatable workflows that turn ideas into polished content: Idea → Research → Script → Visual Storytelling → Narration → Motion → Distribution Most people see the final video. The real leverage is in the production engine behind it.

I’m under the hood leaning how to make my code motion designs better. Taking it multiple steps at a time and iterating over each of the specifics. It’ll be better and theories will be better explained. But in the mean time how does this look?

Everyone wants AI agents that can do things for them. → Trade. → Pay subscriptions. → Manage portfolios. → Execute workflows. → Interact with apps. But there's a problem. The moment an AI agent can spend money, you have two bad options: 1. Give it your private key. 2. Give it unlimited approvals. Both are disasters waiting to happen. So we built ForgeOS. ForgeOS is an onchain operating system for AI agents that lets agents act autonomously without ever controlling your wallet. Instead of giving agents access to your assets, you give them permissions. Those permissions live onchain and are enforced by the blockchain itself. Here's an example: You approve a maximum of $500 USDC. A portfolio agent receives permission to rebalance assets. That agent can create a second agent to handle payments. But it cannot give that second agent more authority than it has. The first agent has a $500 limit. The second agent might get a $100 limit. Trying to increase those permissions simply fails onchain. Not in the UI. Not in our backend. Onchain. The blockchain becomes the spending firewall. We also wanted agents to be able to think and pay for themselves. So every decision runs through Venice AI using zero retention inference. If the AI request requires payment, the agent pays for the inference itself using USDC on Base through x402. No API keys. No subscriptions. No credit cards. Just pay for exactly what is used. Everything runs through a MetaMask Smart Account. Every action is gasless through 1Shot. Every transaction is confirmed through signed webhooks. No polling. No fake confirmations. No optimistic states pretending something happened when it didn't. And if you ever want everything to stop? One click. revokeAll() Every agent. Every permission. Every delegation. Gone. ForgeOS is built on a simple belief: AI agents should be able to act on your behalf. They should never be able to act beyond your authority. Live: forgeagent.vercel.app Demo: youtu.be/xDsXfm0Iyhk

Tell us why’re you up this late?