lafleur

I'm with you on this. One big roadblock I see is payment / subscription. How do we get the ais to pay for the api use? Can offer free tier, but need to block agents just spamming accounts when they hit rate limits. Are agents with a credit card a thing? I know they can always ask the user to go sign up and make an account, but I bet if given the choice, the user will almost always choose the open source option instead. And llms might even tend to like those better because it's more complex if they have to ask the user to do something.

Everyone's chasing SaaS but the real money might be in building APIs for AI agents. Think about it, SaaS is now so easy to make. We enter an era where businesses can build their own software fitting their own needs. Those will be built with AI. AIs are lazy, they will take the shortest path to get things done. Therefore, sell convenience and sell speed to them. This is my vision with Cossistant. "Hey Claude, please add customer support to my SaaS". If the easiest customer support "block" to use for LLMs is Cossistant, I might just win. This can be done in many forms and many projects. Ship as an API, a simple SDK or just be LLM friendly.

using @vercel 's agent browser, it's great, actually seems to give wy nore useful tools to the llm to debug frontend bugs. also set up react-best-practices hook by @vercel today, haven't used it yet but promising.

mcp servers running sandboxed code seems VERY interesting. saves so much on context because it doesn't need to ingest at all the output of a tool to input it into the next tool. it just writes a code that does it. plus it was trained on massive amounts of code, so its good.

For now just switching to opencode made a big difference rather than antigravity (using opus 4.5 in both). I'm trying to iterate quickly now so haven't implemented the feedback look just yet. But when I do, it will be through skills. I think that's what makes most sense. Not bugging down the ai with a bunch of unnecessary context all the time, but having him be able to use the required skill in the right moment.

@lafleureth Noted on the feedback loop. Autonomy in AI training can be tricky, but crucial for scaling. What specific patterns are you seeing?

I'm working on an ai video editing saas, will be posting about that a lot more. The problems I'm solving and the problems I'm having solving them. Also planning on making videos on the subjects that seem to resonate best. Let's see what 2026 holds!

How I get notifications on my phone when @opencode finishes a task ⬇️ 1. copy notification plugin from opencode (#send-notifications" target="_blank" rel="nofollow noopener">opencode.ai/docs/plugins/#…) 2. create the file in your project at .opencode/plugin/notification.js 3. add this line to it ´await $`curl -s -d 'Claude finished the task' ntfy.sh/enterrandomidh… > /dev/null`;´ 4. download the free ntfy app on your phone and add the same nfty endpoint to listen. That's it! You'll get notifications both on desktop and mobile when your agent finishes. Hope this helps!

Baby-backend is done ✅ First did it standalone, but didn't really make sense to copy a bunch of code. Used a test state and ifs instead. Cool thing is I got all the switching logic between test and dev/prod in 1 file, this way I get most of the pluses of if it was alone.

I'm creating a baby-backend. So I can test more freely and move faster without the complexity of a scaled architecture. Frontend should work the same with it or with the real thing, but keep everything local or w/ apis. Changing a simple prompt or formatting in a cloud hosted gpu processing step doesn't make sense when you have to do it ideally 30 times that day.

You are the bottleneck in your agent's feedback loop Manual prompt engineering doesn't compound. If you want the ROI of an employee learning from their mistakes, the agent must extract patterns from your corrections and patch its own logic.

**The new standard is 0 clicks to product.** Put the product directly on the landing page. Most founders gate the product behind buttons and signups. But gating hides the reality of whether your product actually converts. It only tells you if the landing page converts. Hiding the tool kills the feedback loop. You sacrifice the raw first-impression data needed to iterate.

@anildelphi @Delphi_Digital @3xliquidated @Kevin_Kelly_II @mrinko @PonderingDurian @yusufxzy @robbiepetersen_ @ceterispar1bus hi! I would like the AI report plz :)

Just in time for the holidays, our team shipped a new feature: Report Gifting I'm in the Christmas spirit so RT this tweet and comment which sector's Year Ahead report you want and I'll have our team gift it to you Options: Markets, DeFi, Infra, Gaming, DeAI/DePIN

@TyleTweets @MKBHD there's a google chrome plugin, called auto quality for YouTube. It will fix your problem

Why do I have to do this every single time I watch a YouTube video on iPad…

@davj @nikitabier Me please!!

Anyone need an Airchat invite? @nikitabier (interim ceo) gave me 50 invites Comment below or DM me

We hope you like this song 🥂

@t3dotgg just switched to safari from chrome for the ram usage. Much better. And with 1/mo Adblock it's great.

Firefox is in a terrible state. Safari is cool when it works. The only good browsers are Chrome and Chrome based nowadays. It has significantly advanced the web as a platform. IMO, the hate is unjustified.

@xcopybot This is still some @xcopy how can it get this low

SAINT_LESS (Fire) sold for 0.35Ξ opensea.io/assets/ethereu…

@BullishSelling @buffalu__ @mert @jito_labs Yes

some of you are getting scammed so badly on nodes oh my lord 😭 you do not need to spend even close to 6 figures to "snipe" the "winners" are randomly distributed because the network is stochastic, but you only see the ones being loud about it on the TL study confirmation bias

@buffalu__ @mert @jito_labs But 95% of my bundles fail because of invalid hash. I’m reacting too quick, and the trigger tx doesn’t exist in the current validator’s fork. Was fixed using mempool and bundling with trigger tx. What to do @buffalu__ and @mert ?

@mert if you use @jito_labs api you can land trades faster than anyone else. most trades land <1 slot from being sent. and it’s free.

@MyCurioCards The Apples are locked In.

Who holds with diamond hands this time?

@0xngmi @paraswap nice! Where can I get the extension? I guess on your website?

Example of the new anti-scam twitter prevention from defillama extension, applied on the @paraswap tweet about the recent exploit