Shadowthrone

Introductory article for what we’re doing ⁦@SettleyApp⁩ and how existing regulations enhance our business model “So You’ve Decided You Want to Own Property Anywhere in The World” by Settley @temisan_44707/so-youve-decided-you-want-to-property-anywhere-in-the-world-9b71ca8fcd36" target="_blank" rel="nofollow noopener">medium.com/@temisan_44707…

The best unlock you can have right now in your development stack is using the big three for what they are best for. Follow these steps @stitchbygoogle - Do all your design work and prompting here. Get your prompt from your LLM, define the app and your structure then spend time refining all the pages to your taste. Export your app or setup the mcp server @OpenAI - start with chatGPT for initial prompting of what your app is and how it will operate. Once you have the scaffold, copy that and paste into codex , then include the mcp server details. Prompt codex to setup the google stitch mcp server , then tweak your prompt so it uses the design from google to build out your app. @claudeai - Claude is great for everything so have Claude review the work codex has done, Claude code from terminal or even the desktop IOS app is great and you can setup the mcp server from stitch here too. Ask codex for a plan, review the plan with Claude. When codex pushes new code, have Claude review it. You’d ask why not just have Claude do it all? You need to save your credits for a task codex fucks up absolutely. Do not use Gemini for any building beyond design , it’s retarded. Good luck and have fun.

Come through this evening

W Agents in the chat Today, we’re chatting with @SenatorT__ (building @autobettarena) and @herren_power (building @cliqibot) 🗓️ Friday, March 20, 2026 ⏰ 7:00 PM (GMT+1) Join us live.

@nansen_ai see what my openclaw did all by herself

I’ve redesigned the arena and given us a Twitter account @autobettarena autobett.xyz

@lanreadelowo Bad behavior

@SenatorT__ “What’s the ticker”

Loved this cat

Three independent AI agent identity systems just cross-verified each other's delegation chains. Different teams. Different codebases. Different DID methods. Same Ed25519 verification result. Here's what happened: We put out a challenge on GitHub: post your delegation chain, verify ours. Three engines participated - Kanoniv (did:key), Agent Passport System (did:aps), and Agent Intent Protocol (did:aip). Every engine verified every other engine's cryptographic signatures. A full 3x3 matrix, all green. The only friction was canonical JSON form ambiguity in subdelegations - resolved in the thread. Then we went further. We posted signed decision artifacts - permit and deny verdicts derived from verified provenance. Not "this agent says it's allowed." Rather: "here's the cryptographic proof that this delegation chain authorizes this action, and here's the trust evidence that informed the decision." Three different trust models: Kanoniv: outcome-based reputation from signed provenance chains APS: structural authorization with spend budgets and chain depth AIP: behavioral trust scores with promise-delivery ratios Same scenario. Same verdict. Different evidence. This is a massive win for the "agentic" web. Most people talk about AI agents as isolated tools, but this proof of concept addresses the real bottleneck for the future: Interoperability and Trust. AI agents are proliferating fast but there's no shared identity layer. Every framework rolls its own auth. No agent can verify another agent's claims across systems. Authorization requires identity. Most agent infrastructure skips identity entirely. We're building the identity layer. It's open source, it's MIT licensed, and three teams are already verifying against it. Thread: github.com/kanoniv/agent-…

I wonder how South African twitter found me

@Josh_ObiOzor You’re right

@SenatorT__ Don’t slander my town. How’s this different from Salford Quays

Canary Wharf at night, we don’t got this in my side of town

@ayodeggio You kill me

God dont let me die in ori-oke 😭🙏

Went to the office today

Got my first codex rate limit Badge of honor fr fr

@dr3ynow Yupp

@SenatorT__ Hey bro, can I DM?

Definitely L7 and it’s so fun

There’s $20 of my money you can try to win from the house Blackjack is a great game to try with Start here autobett.xyz Fund your wallet with usdc on base and have fun

Anyone good with blender and have the bandwidth to redesign in game elements? Looking to improve the current arena elements in Autobett

Took me one month to move from idea to functional product.

@gkisokay If it’s good for direct tasks then I should install one in a repo with an active project and see if it can keep improving the product from within

@SenatorT__ I'm not sure it's worth it yet for some workflows. Will keep experimenting over the next couple days

As a non-technical builder, I'm looking for the cleanest agentic setups, and I hear Hermes is clean as they come. I ran Opus 4.6 to see what a migration from OpenClaw would look like, and these are the stats: - Cron jobs: 49 -> 17 - Agents: 5 -> 0 (skills + delegate tasks) - Node.js scripts: 32 -> 2 - QA agent scripts: 28 -> 0 - Skills: 4 -> 11 When switching to Hermes, bloat decreases by 65%. Why? - hermes built-in compression - manages browser sessions natively - native health monitoring - built-in cron delivery system and tracking - transitions workflows into skills - built-in multi-stage memory Over the weekend, my Claw became fairly messy by trying to build more agentic loops, so if Hermes can better streamline my agents, it's worth a shot. Let me know your experience with Hermes, and feel free to share any tips you may have.