Raphaël T-M

@wzich Working on it!

@raphaeltm_ brother this scrubber

Where I was. What I was building. I love building like this. #ai #agents

who's launching a dev tool on Product Hunt this month? happy to help and support

@wzich I knooooowwww 😂

@KaranVaidya6 @gdb @garrytan I broadly agree. The one case where I prefer mcp (or really any kind of self documenting API) is when you have limited ability to install things on the system. Then mcp over http is nice.

Okay, @gdb is team CLI all the way. @garrytan thinks MCPs suck. So we hit the streets of SF to see if the city agreed. We posed a simple question: MCP or CLI? - Basically everyone under the age of 35 said CLI - One person said MCP was as bloated as Java - & unsurprisingly, numerous people told us to touch grass Final score- MCP: 3 vs CLI: 17 SF has spoken, and @composio listened. Our universal CLI is now live! Drop your best CLI vs MCP hot take in the comments and we'll send the best ones some very sick gear 👀 Link to try our CLI in the next thread ⬇️

@tokifyi Woohoo!

only ppl i like can reply to this

We’re excited to announce the launch of Gencove Consumer! $49 for a sequencing-based genetic baseline. Built for individuals who want to better understand their health and for a future where genetics is standard in every consumer health product. 👇 enterprise.gencove.com/blog/introduci…

Resource usage looks pretty low here and could probably be more efficient, but I find it's really spiky: if an agent is just reading files, it's all good... but if multiple agents on one VM suddenly decide it's time to run 1000 tests, then it's... not ideal.

It sometimes feels really stupid to build like this, but it's also really satisfying? I just kinda braindump into SAM, and when things look half decent, I tell it to dispatch the stuff I think looks good... and then off it goes. #ai #agents #coding

Would be nice to get the power of AWS without having to touch that console anymore, no? #aws #defang #cloud #deploy

Alright. As promised, @StevePaolasini and I have been really busy over the past few weeks compiling the full Canadian Startup Visa (SUV) program backstory complete with the fraud and mismanagement facts. And damn, do we have a story for you. Stay tuned for the full article when it comes out but let's start with some fun facts we discovered in the process. Like many things that have gone wrong, the SUV started off with great intentions. An ambitious, unique, innovation-driven pilot program meant to replace the antiquated Immigration Entrepreneur Program that got phased out in 2011. SUV had a very straightforward objective: attract innovative founders who would build companies in Canada and contribute to its long-term economic growth. What made the pilot unique was the selection mechanism: the private sector evaluated the business idea first while the government assessed admissibility second. In practice, this meant that before applying for permanent residence, applicants needed Commitment Certificates / support from a designated Canadian organization such as: venture capital funds, angel investor groups, or business incubators. The SUV pilot was launched in 2013 and the IRCC decided to convert the program into a permanent one in 2018. It had great initial results! Lower operating costs than the previous entrepreneur program and applicants raising higher capital in Canada: reinforcing the idea that private-sector validation was working. However, IRCC's own evaluation of the SUV pilot indicated a very important weakness: the government had limited visibility into the ongoing activities of the designated orgs. This is important because that's where all the issues started. IRCC's follow-up, 2023 program evaluation report indicated that "One-third (33%) of surveyed clients reported “an opportunity to immigrate to Canada” as the most appealing aspect of the SUV Program." The report also pointed out that some designated orgs were allegedly charging applicants additional fees to assess their businesses or create fraudulent documents and immigration applications. Let's talk about some of the biggest offenders. Starting in 2019, Empowered Startups was featured in a series of Federal Court decisions revealing troubling arrangements involving their applicants. There are multiple, public cases documenting that applicants each paid this designated org 300K CAD in incubation fees! Another one known as Manitoba Technology Accelerator (MTA) operated under two different names and submitted HUNDREDS of applications under both of them over 2023-2024. The total applications submitted by them over that time period was upwards of 1K cases. Funnily enough, MTA only lost its designation temporarily in 2025 and Empowered Startups never got de-designated. Instead of dealing with the fraud-abetting organizations, in December 2025, the IRCC stopped giving out SUV open work permits. And in January 2026 they indefinitely paused the entire SUV program. These drastic actions make sense from their end. Their backlog is now over 45K people and there are only 500 (!!!) spots allocated for business immigration in the 2026 levels plan. That is close to 90 years of inventory, not something that is feasible or even realistic to deal with... The main questions now remain: 1. Why did we turn a blind eye to this sheer scale of fraud going on in the SUV program for years? The warning signs were there as early as 2019 but the program only got paused after 2025 2. Why have these organizations not lost their designation before? Why did MTA only lose its designation temporarily? Why is there no further investigation being done into these fraudulent activities? 3. And most importantly, what on earth are we going to do with a SUV backlog of 45K applicants with support from predominantly questionable organizations and close to no spots available now for business immigration applicants? Is this where the powers of bill C-12 will potentially come in? I'm not sure anyone has the answers right now. But we do need to deal with the consequences of this mismanagement before launching a new program...

Defang CLI v3.3.2 released. Fixes some incompatibility issues with the old Buf/Connect Go modules and our new iter.Seq based logging infrastructure. I suspect Buf did not like getting invoked from the coroutine that got created by iter.Pull.

@LioLunesu I'm pumped about some of the upcoming changes and how we'll be able to address this kinda problem.

Defang CLI updates are pretty straightforward: our GitHub page shows the changes in each new release. One thing we struggle with is how to communicate changes to our "CD task" code. This is the Pulumi code that provisions the Compose infrastructure in your cloud account. Right now, Defang always uses the latest version of that code. This has enabled us to quickly iterate on features and get quick feedback. But it has resulted in some customer issues when the code regressed or it caused an unexpected resource replacement in the cloud. Going forward we can no longer afford that. So starting from the next version of the CLI, the CD code will be "pinned" per project, ie. if the project was first deployed w/ version X, it keeps getting deployed with version X. So now we need a new mechanism to inform users of new CD code versions and change-log, and opt-in to updates.

@karpathy @nummanali I feel like building on top of cloud infra primitives to manage tasks and information is the way to make things really scalable, then allowing agents to communicate back with that infra over APIs (exposed as MCP or otherwise) to manage their context, or delegate to other agents.

@karpathy @nummanali This feeling, and the desire to manage multiple agents from my phone with my voice specifically, is why I started building this thing: simple-agent-manager.org I kinda think of it like the k8s control plane, but for agents. At least that's where I'm trying to take it.

Claude Code teams with tmux is really cool When you run with team mode enabled in tmux, it automatically opens the additional terminal in pane I don't really get my main agent to orchestrate, I chat to them myself CLAUDE_CODE_EXPERIMENTAL_AGENT_TEAMS=true claude

@DefangLabs @DefangLabs abstracts that away in a way that is usable for the vast majority of apps, so it works locally or in different clouds exactly the same way. If you need to break out of Defang, you can because you still control the infra.

@DefangLabs Normally you have to deal with this: you make an application arch change, then you need to reflect that change in your infra definitions across each cloud, and need to reason separately around the infra primitives that allow you to replicate that arch consistently.

@DefangLabs makes this so much easier. Just one compose.yaml file and deploy to any customer cloud account.

I'm finding this really useful. I find different agents useful in different contexts, but they all roughly work the same way. It makes a lot of sense to keep communication with them consistent with a common protocol. Have any of you built on top of ACP? agentclientprotocol.com/get-started/in…