@aidenybai You can group only the tools you need into a single MCP with app.mcpmarket.com
English
Graeme
340 posts
Graeme
@itsgreyum
PM by day, building at night. https://t.co/seK5MFrlBI
Canada Katılım Ağustos 2020
727 Takip Edilen97 Takipçiler
So I guess this was just a lie since they ended up pivoting
conor brennan-burke@contextconor
PSA: we’ve onboarded more customers than we can handle, so we’re pausing new demos to focus on current customers if you’re booking now, we’ll reach out in december once we come up for air thanks for the patience everyone
English
every AI company charges the same per token whether it's 2pm on a Tuesday or 3am on a Sunday.
their GPUs are not equally busy at both times.
electric companies started dynamic pricing 50 years ago. pay less at night, pay less on weekends, pay more during peak. Obvious no?
if Claude or GPT offered off-peak rates, every company would shift their batch jobs to nights / weekends overnight. literally.
the compute is sitting there. the capacity fluctuates. the pricing doesn't.
this feels like an obvious unlock that nobody's building yet.
am i missing something?
English
Talking to smarter folks than me, I'm convinced many of the AI folks in my timeline are full of shit.
Nobody is "running 20 agents over night" and building stuff for actual users. Maybe some are building internal tools or disposable software. Maybe.
But building software people like using? That doesn't get hacked on day one or blow up after the 3rd user? Nope.
I don't even understand what that's supposed to look like. Do you work out a 57 pages document that perfectly describes what you want to build and then summon 14 agents and have them run wild for 6 hours? And what comes out on the other end isn't a broken pile of shit?
Nope. Not buying it.
PS: it may also be that I have an IQ of 82 and can't figure it out.
English
@icanvardar Theres more usecases than just coding. Gemini flash is great. Cheap and fast workhorse for agent workflows.
English
@DamienWayne Unless you absolutely need it, just focus on product/distribution. There comes a point when the numbers speak for themselves.
English
Day 20 of my official fundraise:
I got into Tiny Fish Accelerator, just waiting on demo day.
Additionally I pitched to Plug and Play which had like 59 ish VC’s present.
Tons reached out, I’ll be honest I am fucking tired of talking to VC’s. I’m about to take the first offer.
Selling to customers was a higher ROI. This is slowing me down
English
@MatthewBerman @conductor_build all the way. Usually ideate/work on a plan in Warp, then move over to a workspace on conductor to build in the background. Just make sure to setup a start script to setup your environment.
English
How are people running multiple agents at the same time? I’ve tried separate worktrees but things still get messy. I want separate environments for every agent thread. How can I accomplish this locally?
English
@ivanburazin That same signal can turn out to be false and you can easily bog yourself down chasing the shiny enterprise deals early. Risky to let 1 big deal define your roadmap.
English
Enterprise validation at pre-seed is usually rare. We had a F500 company saying, "we'll pay you to build this out."
That de-risked the investment significantly.
What we were building was no longer a devtool with unclear monetization. People started seeing it as enterprise infra with a designated buyer.
The strong intent from key decision makers gave angels confidence that we weren't just building for developers.
If you can get some early enterprise signal (even if it's just an LOI or pilot commitment), it changes the entire fundraising dynamic.
English
@hwchase17 Better MCP client support for prompts/resources (plus ability to inject as user msg rather than tool result) could make MCP the ‘backpack’ for portable agents and skills. Take them with you to any platform, plug n play.
English
@adriannalakatos 👋 building management layer for teams using agent skills and MCP. Mcpmarket.com as distribution (600k monthly uniques), 5k waitlist, 3k skills downloaded a day. Been too focused on shipping to package an application together.
English
know a cracked builder? drop their @
finalizing invites for canopy. got a few slots open for interviews today + tomorrow. first decisions out EOD friday. april 15 start.
100 teams irl. 400 online.
hardware. software. media.
investing $100k - $250k checks
English
@kiwicopple @garrytan @PostgreSQL @pgvector VFS with a Postgres store and you’ve got the best of both worlds.
English
neat setup: @garrytan started storing all 13K markdown file of his personal knowledge base in @postgresql + @pgvector
a personal search engine for everything: people, companies, meetings, ideas, media
Garry Tan@garrytan
I'm open sourcing it MIT license so we can all speed up and have our own personal mini-AGI. It's been amazing for me and I want you to have it.
To install GBrain in your OpenClaw, just paste this image to your OpenClaw or paste this text:
Set up gbrain (github.com/garrytan/gbrain) as my knowledge brain.
1. Make sure bun is installed: curl -fsSL bun.sh/install | bash
Then run: bun add github:garrytan/gbrain
2. Run: gbrain init --supabase (follow the wizard to connect my Supabase database)
3. Scan ~/git/ and ~/Documents/ for markdown repos, pick the best one, and run: gbrain import
English
This would be amazing. Currently hacking it by using prompts for user invoked (but client support is week) + a search_skills / get_skills tool call for agent invoked, but being able to inject as a user message would be great.
A ‘skill’ tool type that gets injected as user message would solve a lot of problems.
English
@RhysSullivan @threepointone What convention are you using, we going to put skills loading into an official MCP extension, and would love to hear what works for you and what doesn’t.
English
Issue is they get delivered as a tool call result where as native skills get pulled in as a user message. Model may see some parts of a skill as prompt injection attempts and not follow as smoothly as it would a user message. Prompts solve this for user invoked, but client support is pretty weak. Overall though, if skills could be cleanly invoked via MCP, you’ve essentially got a pluggin knowledge hub you can use across providers
English
@harjtaggar The more decision making that gets delegated to agents the less we actually need chat. Chat becomes the equivalent of ‘hopping on a quick zoom to clarify requirements’. Most things can be done via defining tasks and letting the agent(s) triage/sequence etc.
English
People underestimate how far chat will go as the right UI because they underestimate how much decision making we’re going to delegate to agents. Looking at a graph or table is going to be something you only do in a code red.
English
@TurnerNovak Not a bad play considering a lot of unemployed people are picking up AI data labeling work on the side while they look for a new job
English
Looks like LinkedIn is building its own AI data labeling business
English
