AI Engineer @Odoo | Founding Ambassador @kimi_moonshot | @FactoryAI Guild Memeber | building agents, tools and sharing what I learn | use Vim/Arch btw
Namur, Belgium
Joined September 2026
- Tweets130
- Following315
- Followers61
- Likes504
Gemmy retweeted
Unfortunately AI has stepped into this weird space where its treated like sports and openai and anthropic are teams and you have loyalties to one or the other and stuff like that which I don’t get, with people that work there being seen as celebrities and such things
Its a weird space right now
Some people complain because of these loyalties
Gemmy retweeted
Excited to be introducing Droid 1.0
On Terminal-Bench, @FactoryAI outperforms all others.
Droid has been one of the most efficient and high quality harnesses out there, and one of the most ones I enjoy driving daily
And actually, they don't stop, and are getting better everyday
Great launch guys!
People will always complain
- Your 200$ will cut in half? Complain
- We're not going to cut it in half? Complain
- Extra API credits on top of your plan? Complain
- Faster and cheaper mid-tier model for more use cases? Complain
That's becoming very annoying
Day 25/
We are doubling all usage limits for all Pro 100/200/500 tiers and Business Premium, effective today.
We heard your kind feedback and won’t be reducing usage limits for Pro 200, so your Work and Codex usage will stay the same. We’re also going to continue offering extended limits for dots for now, and usage of your dot will still continue to be separate from your default usage.
And we'll of course also keep shipping better models and other efficiency improvements that increase your effective usage over time.
Gemmy retweeted
Kimi now works with Gmail, Google Drive, and Google Calendar.
Summarize Gmail threads, find files in Google Drive, and plan your day with Google Calendar — with secure OAuth access you control.
Try them at kimi.ai/plugins
100% agree,and I also come back to Fable for creative design
a tip I use for design workflows is that I tell Opus 5.5 to have a Fable subagent as a creative director and it really works
Fable has something different with creativity, even if Opus 5.5 is better overall
I find myself always coming back to Fable for planning.
These days, I'd say I do about 80% of my planning with Fable 5.1 Medium.
While Opus 5.5 has made its way into my execution layer, I still like Fable more for planning.
I've been thinking about how to quantify this, much harder to build eval suites around things like planning than things like agentic coding performance.
Building eval suites to determine how well a model solves coding tasks has a lot more determinism to it, and is much easier to assign very clear numbers to w/respect to scoring.
Planning in many ways is subjective, but I can't stop thinking about ways to build evals around this.
I know for me, I definitely like Fable more than Opus for planning. I, as a puny human, have run dozens of tests, and I just like Fable more, it's so clear to me.
But I don't know quite how to build an eval suite to replicate what I'm seeing myself as the difference. Still noodling on this, not much more to say for now.
Anthropic's new guide on folding subagents back into one agent loop is worth a read
An interesting idea is that you should split agents by context, not by org chart.
That matches where I landed. The only agents I'd keep separate are the ones that need a fresh context: a reviewer that never saw how the code was written, and independent research paths. The rest is worth testing as skills in one main loop
claude.com/resources/article…
That's literally what I said about Terra
it has its advantages, and if you're powering customer-facing products, you'll find it very handy as an option
Coding tasks are not everything
People want every model to compete in every direction, which I think is unrealistic
There are different tiers, so you pick what fits the task. I've been using Terra for different tasks all day today, and it's been fast and good for those use cases
It won't be the right fit for everything, but it's been useful for me in what it's marketed for
Honest GPT 6.1 Terra review
It's fast for computer use that's it, anything else it's not really useful for me, also the model stops working a lot for no reason like I give it a task and it completes half of it and stops to tell me i did X and Y and X and Z left like why just keep working and finish the task that does get really annoying when you just wanna AFK a task.
I mentioned other usecases here
Gemmy retweeted
Kimi Ambassador になりました！
Thank you to the Kimi team at Moonshot AI for this opportunity!
With WebMCP, you can do too many amazing things. You can let your agents play games with you in the browser, like this family game spymission.dev
You just need a bit of imagination and to build stuff you enjoy!
That's unexpected, pretty excited for this one 🔥
Going to drop a detailed practical review about it soon!
really love what HyperFrames is doing
seems After Effects quality right? but it's made with @HyperFrames_ Studio in pure and raw HTML!
how it was made was pretty simple, choosing some inspirations from our inspirations library and boom, 11 mins with Opus 5.5 and I had this video done, fully editable HTML
Gemmy retweeted
I designed all of these landing pages in the last two months, using @higgsfield , Claude, and Cursor with my custom AI designer skill.
The skill is open source: github.com/elayadesign/ai-de…
