Reese

The @SkaleNetwork Docs v3.4.0 are live! > @corbits_dev facilitator is officially added > New easy-to-follow guide for @relayaisolana > Updates to @GoKobaru facilitator page > Step-by-step Testnet Tokens page (including how to use the Circle faucet and Portal bridge) > BITE Sandbox is now easier to find > Fixed a couple of bugs docs.skale.space

x402 Weekly Wrap and Thoughts #16 A wild week in x402, lots of thoughts and ideas below. > The (I think) biggest news of the week was probably @stripe dropping Machine Payments bringing x402 to Stripe merchants. Plus they dropped official MoR support which is huge. > @dexteraiagent is out sending emails to people like @cuysheffield . The relevancy is that AI delegation and capabilities is enhancing. AI accelerates AI accelerates AI. > @SkaleNetwork @googledevs @CoinbaseDev @edgeandnode and @virtuals_io have seen a significant number of incredible teams submitting Encrypted x402 Applications to the SF Agentic Commerce and x402 Hackathon. Still ~13 hours left if you want to get hacking. > Four more big companies added direct support as x402 merchants this week: @hyperbrowser (YC24), @coingecko, @AlliumLabs, @Quicknode > @framesag @CoinbaseDev @LitProtocol @privy_io and a bunch of other teams have dropped or shipped new agentic wallets in the last week > @mrdn_finance shipped Mpay, bringing multi-token transfers to a few different networks > @daydreamsagents is setting the standard for agentic search with xGate > @programmer hopped on SKALE School and shared some incredible thoughts on where x402 is going as a standard and what he believes developers and builders should be shipping > @relayaisolana and @GoKobaru both added support for USDT, WBTC, and WETH which are all EIP-3009 compatible on SKALE Base Mainnet and Testnet opening up opportunities for agents earning or transacting in non-USDC tokens > @relayaisolana is the first facilitator to add support for encrypted x402 transactions on SKALE Base with BITE Protocol. No action needed on your side; just starting paying on SKALE Base with RelAI and you are picking up a couple precious seconds of privacy in a world where agents and bots are analyzing everything for an edge My biggest thoughts this week are: > The opportunity as a merchant is massive. Many x402 resources are not actually working and it's an open market. There can (and will be) dozens or hundreds of the same things just like in the real world. Start building. > Build for where every is going. Don't ship ANOTHER agentic wallet. Don't ship ANOTHER OpenClaw. Think about what agents need to fund their wallet. Think about where lobsters want to go hang out. Think outside the box. > The first facilitator who targets being a Merchant of Record with refunds could dominate > More agents means more congestion onchain. This means chains with local fee markets can see much higher prices if everyone is using USDC. Chains with monolithic architectures don't have a way for agents to quickly acquire their own blockspace. Make sure you deploy your x402 endpoints and agents on chains that don't penalize for growth like SKALE. > Integrating x402 outside of Base and Solana still has extra hoops to jump through as they aren't the primary targets for CDP. I'm building out armory.sh to solve this ASAP My biggest belief right now is that anyone involved in x402 is still early and there is so much to build. So many incredible teams and people are involved. Not sure where to start? Shoot me a DM. Want to see your project listed here next week? Shoot me a DM or tag me when you share it. Want to share your project now? Leave a comment below. Thanks! > A reminder that none of the above is legal or financial advice. None of my tweets ever are.

x402 Weekly Wrap and Thoughts #15 Apologies for missing last week. Have been heads down building + helping developers integrate with SKALE. Wrap first, then thoughts (make sure to read to the end for my deep thoughts). Prefer my thoughts over updates? Scroll down past Key Updates and checkout what I've been focused on the last couple of weeks. Key Updates - It's no secret that @openclaw is arguably the hottest news of the week (besides the broader market panic and confusion). The closest thing we've seen yet to full autonomy. Excited to see where this goes, but there are quite literally to many things to call out. - @samrags_ brought Sign-in-with-X to @coinbase x402 v2. This brings a huge upgrade in pre-paid power to x402. Great example from @fabdarice -> pay a one time $10 credit and then every subsequent x402 request identifies you and decreases your credit - @CloudflareDev have continued shipping (shoutout @williamallen ) with a brand new update to their x402 template on Cloudflare - @SkaleNetwork has had a flood of announcements and in my (very biased) opinion continues to show it's strength in the agentic era with a slew of partner announcements, new teams going live, more open source contributions and has partnered with @CoinbaseDev @google @virtuals_io and @edgeandnode for the upcoming 100% hybrid x402 and agentic commerce hackathon which includes IRL co-working space sponsored at The House of Web3. Register now, it starts the 11th: dorahacks.io/hackathon/x402… - @corbits_dev Pontus joined me for an x402 Roundtable yesterday and dropped some awesome alpha around some big updates coming to Faremeter and Corbits. Excited to see them continue to push the pace on x402! - @ethereumJoseph spoke at the LFDT Member's Summit and shared thoughts specific to x402 stating (slightly paraphrased) "that it will enable agents to pay one another" using x402 (h/t @HashgraphOnline for the tweet) - I randomly stumbled across @fere_ai which looks awesome (can we chat team?) - @awscloud dropped an incredible guide on monetizing an application with x402, Cloudfront and Lambda@Edge: builder.aws.com/content/38fLQk… - @HoumanShadab continues to impress with API failure insurance for x402 agents on @NovaNet_zkp . Pretty sure someone -- cough cough -- @iFalcore told me this was a dumb idea ~2 months ago. (I said otherwise). - @OnchainLu continues to impress with some of the best statistics and reasoning in the x402 space. The breakdown on what's real vs not was pretty amazing and really shows the value in having "real services" and not gamed txs at this point in time. - @Superfluid_HQ build the x402-sf extension which brings streaming to x402. (Can we connect team?) - I shipped armory.sh, a batteries-included x402 SDK in early beta designed to help me better explore and understand how to abstract the complexities of x402 +++ and so much more, if I missed you leave a comment below with what you shipped, did, or talked about this week 👇 Thoughts No 1 - Agentic Discovery is Fractured Not all facilitators in x402 are supporting discovery (from what I can see) and not all agents will be registered on all chains. This creates a simple but clear problem: agentic discovery is dynamic and complex. I originally believed that discovery would occur at the facilitator level and be a deep moat for them, but am realizing more and more that it's probably going to be a layer above them. No 2 - Creativity is Hard This is a funny one to think about but I think a big problem for a lot of engineers is the lack of creativity. It's really hard for someone who is used to be told what to build to look at a blank canvas and imagine something world changing. I think it was from the Silicon Valley TV Show sometime when they were building Nucleus. They called them "dreamers" or something like that. Funny enough I think those people will be more and more valuable. I don't think this is an "all SaaS disappears" issue or a "engineers will go extinct" problem. I just think the value shifts very heavily toward people who don't just see a vision -> but use the tools on hand to execute. No 3 - Guardrails is the "Unsexy" Winner of Agentic Commerce Seeing a lot of teams and projects and ideas spin up focused on how to give an agent money. This is going to be in my opinion an absolute bloodbath of screw ups and lost funds until we get it to click. There are just so many tradeoffs and factors to be considered but the reality for me is that ungoverned "autonomy" that can be steered doesn't play well with things that should stay private. Ben exploring some projects like @LitProtocol on this and excited to see where it goes. No 4 - ERC-8004 will grow in communities ...primarily across micro-blockchains. I've been thinking a lot about the value of discovery and finding what you need (back to No 1). Specifically in relation to ERC-8004, I could see groups of agents buying their own blockchain autonomously via SKALE (yes this is possible) on @base via @SkaleNetwork Manager and having their own mini guarded deployment of ERC-8004 that is hyper-curated toward their goals as an agentic group. This is a reminder that all of my posts are for informational purposes only and is not recommendation to purchase anything or legal/financial advice of any kind. If I missed you in the recap, leave a comment below with what you shipped in the last two weeks and shoot me a DM!

Good differentiated SaaS is fine. It takes people resources to build and manage and maintain. The small companies who are subsidized by the enterprise contracts shouldn’t be the concern. The concern should be for SaaS companies with a razor thin moat and little to no differentiation from their competitors.

I don’t think people want to build their own apps or software. I am building in the contrarian perspective of this: AI is amazing at a lot of things, but it’s not easy to use. This needs to change.

If your trying to bring AI Agents into your business or enterprise; you need to come to grips with a few things first: 1. Replacing workflows End to End is hard, especially for companies that have undocumented processes 2. Every company or department has that one person who handles a lot of things ans wears many hats. Don’t force AI on them, work with them to see how AI can help 3. Work with a consultant familiar with deploying internal systems for automation with or without AI. The reality? Not everything needs AI. DM if you are curious how agentic systems and solutions could help your business.

@initflorian Agree fully

@theagenticdev docker compose that just works is table stakes for dev adoption

If you are building an open source AI tool with a cloud-option; a developer should be able to go: a. Locally hosted Docker Compose file that "just works" b. 0-1 self-hosting with Railway (officially maintained deployment option) If you don't have at least the first option, you are limiting your potential developer adoption.

A major issue that I'm consistently facing is the lack of cohesion between tools. A lot of AI SDKs, frameworks, and tools are very strong at the surface but as you dive deeper are very much so vendor lock-in, lacking substance, incomplete, etc. This is the nature of anything really. Some areas I've been focusing: a. Rolled my own UI SDK to work with Vercel AI SDK frontend. I wanted the experience for some one interacting with the agent to be very specific in terms of how they see what he AI is doing b. Creating a two tool semantic registry. This is not working for some reason (yet). I have the tools, I have the embedding pipeline, I have the tool; it's just not selecting the right tools (could be the model, could be the prompt, could be the tool) but the goal here is that a single agent can act as a router to many other agents and tools with single search & execute tools c. Artifact generation needs to be more opinionated. I think the concept of artifacts is under utilized and I'm sure Anthropic and OpenAI will improve over time but the lack of auto creation of documents and information that can be consumed outside of code and images is a major issue. Sure pushing to 3rd party systems is more common (from what I've seen) but then that reduces the context and memories in the platform and reduces the power unless you have EVERYTHING hooked up for search

Spent a lot of time trying frameworks over the last few weeks, will be doing a detailed write up over the next couple of days but it’s readily apparent that the majority of SDKs are are barely past PoC mode and struggle with anything “out of the box”. Huge opportunity for a multi-agent UI elements SDK that actually works.

@ichebykin Trying this ASAP

Cursor and Claude often get libraries wrong. Hallucinated methods, broken generated code, and wasted hours fighting with the agent to fix it. We want to solve this - we created a web wizard for generating integration specs for coding agents. Simply connect your repository, and choose a library. We will scan everything and prepare a plan that can be executed by any coding agent. Try it out at: sourcewizard.ai/planner Oh, and we're fully open-source: github.com/sourcewizard-a…. We don't store your code and you can verify that.

@EyalToledano DMd, looking forward to testing it out!

the PRD is becoming a coordination layer, not a stale doc pm used to be: roadmaps → handoffs → status updates then: generate plans in seconds with ai now: plans built collaboratively with global context, aligned live, executed by agents without drift dm me for early access

@austinh___ What are the biggest problems you need solved right now?

I borrowed a key lesson from my Ramp days: hire people for their trajectory, not their past experience or brand names on their resumes. Raw talent, big ambitions, and the willingness to learn quickly. It’s a magical combo.

Agentic coding models have some major DX issues with Typescript: 1. dynamic imports 2. IIFE 3. Lack of ability to figure out edge cases on features (just implements them in a way that breaks ALL the code Are they amazing? Yes. Curious to see if this can be solved via improved context engineering via the harness or if just really requires constant updates to the models so they can be trained on the edge cases of public npm packages.

As you build to solve one problem, you uncover dozens of others. Solving each problem results in a lot of code.

Operational Agents

The @cursor_ai tool call for create plan has been erroring out all day :(

The pickaxes and shovels of the agentic era: - Power - GPUs/Cloud Compute/Bandwidth/Data - Search Tools - Memory/Context Tools The gold: Currently coding agents, subject to change

A couple of companies building AI Search right now but a major issue is because there are only 2-3 active working ones; the cost is so high it’s infeasible for basic LLM grounding.

Some models are capable of handling directly, but often you can extract info first and reduce the cost of the LLM call. This is one of the things I’ve build recently — pre and post extensions for the LLM. Why? The cost of manually parsing a PDF and using a text only LLM seems to be cheaper and faster on average than a multimodal LLM capable of parsing directly. What do you mean read PDF? Extract to markdown? Summarize?

I’m surprised there’s no MCP for reading pdfs