Agents. Applied AI applications.
Delaware
Joined January 2024
- Tweets663
- Following125
- Followers483
- Likes497
Inkstone: the secret weapon for knowledge work.
Write. And AI helps you edit, research, fact-check, proofread.
AgentSea retweeted
It's 2030. AI didn't kill jobs.
It killed the $300 million budget. Now a 5-person team in Jakarta is beating Marvel.
That's not dystopia. That's abundance.
Mobile alpha coming today. Works sweet on the phone. Unlike other apps that lose all their advanced functionality on mobile, we keep all the features but adapt the interface.
React in web browser for now and app store coming later.
But it's cool!
Without you, AI is nothing. You're the idea machine. The dreamer. The doer.
Inkstone puts AI where it belongs, in the passenger seat.
You write and AI helps with the research, checks the facts, proofreads, edits.
No more 50 tabs. No more copy pasting. No more juggling.
AI is so much better when it's a co-creative experience.
Especially with writing and thinking and brainstorming.
Inkstone is a co-creative writing app that lets you deep focus, but with powerful background agents that can proofread, fact-check, deep research, web search while you stay in the flow.
inkstone.pro
Inkstone is the co-creative writing app that lets you deep focus, while giving you powerful background agents who can proofread, factcheck, deep research, web search while you stay in the flow.
And when you need it, it can edit docs like Claude edits code.
inkstone.pro
What if you had a rich text editor, deep research, web search, proofreading, fact-checking and code agent like precise text edits in one?
One app instead of $400 in subscriptions and copy pasting like crazy between them.
Coming January.
Waitlist -> inkstone.pro
AgentSea retweeted
For months, I've been quietly building a prototype of something just because I want it to exist.
Papyrus is a word processor, editor, proofreader, fact-checker, deep researcher, brainstorming partner, all in one. It takes your rough draft and helps you skip three revisions.
Now I'm considering pivoting my whole team to build it out, but first, I need your vote. You email is your vote.
LINK IN COMMENTS.
I don't even care if you use some email you never look at and just use for sign ups. To me one email is one vote and says, go ahead and build this damn thing out and make it awesome.
I built it because today's apps are the bloated dial-ups of writing (Word, GDocs). They've got a hundred freaking buttons I don't need and bury the ones I do need. Or they're AI marketing slop toys that promise to read my mind and pump out garbage. They just don't get the real work of writing and they don't get me.
I want a clean, focused space with a real AI co-pilot.
I don't need it to do the all the writing for me, just like I don't need Claude Code to do all the code for me.
I want to work with it. I want a writing partner that acts like a full writing team on-call 24X7: proofreading, fact-checking, and running deep research in the background while I focus on the hard parts.
I'm tired of cutting and pasting between a dozen tools so I started writing this thing.
But won't the big guys just build something like this?
Sure.
But it will just bolt your ass into their ecosystem.
Google Docs will force Gemini on you. Whatever office suite OpenAI pumps out will only use GPT. I want to use any model. Open or closed. They're commodities. The app is the thing. I want the best tool for the job, GPT-5, Kimi, Qwen, Claude, whatever comes tomorrow. I want a fluid, flexible workspace.
I'm building this thing for people who've got critical thinking and who build, who don't want to outsource every damn thing to the machine. So if that's you and you believe this should exist, I really need your vote. Tell me to build it and I'll go all-in and make it a reality.
Thanks for giving me a moment of your time.
AgentSea retweeted
I got super tired of Claude Code getting amnesia 🧠✂️ after every auto-compact.
So I fixed it for real.
Meet Flashbacker 🧠⚡: github.com/agentsea/flashbac…
Install it in any project you're working on with 'flashback init' and watch it go!
It gives Claude much better memory by fetching context from previous conversation history (stored in ~/.claude/logs/), as well as some custom files stored in .claude/flashback/memory/ REMEMBER.md and WORKING_PLAN.md
Use /fb:working-plan to keep that plan updated and /fb:remember to store precise, simple rules that Claude should never forget to avoid repeated mistakes that it does over and over.
It works awesome. It shrinks the context window a little but gives Claude a much better understanding of what came before. The best is that it fetches and loads up a sanitized version of the last 10 entries in the last two conversations, as well as deep plans about what to do in an environment.
I also created callable sub-agents that get fed previous context callable with @agent-{AGENT-NAME}
Many of these agents come right from the awesome Super Claude project: github.com/SuperClaude-Org/S…
Lastly I built an agent 'discussion' that calls multiple sub-agents and then summarizes their output and uses a debate moderator to outline what each agent recommends.
I also built some agents that do heuristic code analysis, particularly /fb:debt-hunter that can auto-detect what kind of code you are running like go, rust, python, node, etc and look for technical debt that AIs are notorious for like TODO, FIXME, PLACEHOLDER, along with duplicate functions and unused functions.
Probably my favorite agent and persona is code-critic, in the style of Linus Torvalds, which is absolutely ruthless in critiquing code, a Dan Jeffries special. 😤
My favorite command is the simplest: /fb:how {SOME_PROMPT}. Basically it tells Claude:
1. What did you understand about what I said?
2. How will you implement it?
This has saved me countless hours. I can usually 1 to 3 shot an implementation by making sure Claude actually understood what I meant and validate its plan before it does it and fucks up my code.
Another awesome one is /fb:hallucination-hunter 🕵️ It ruthlessly looks for completely bullshit, made up AI code that is broken or fakes doing something useful.
Most Claude "plugins" and "apps" are basically just a collection of prompts. That's useful but not very powerful.
This is pushing the limit of what an app is in Claude, since Claude really does not have true plugins. I use namespaces as best I can to not shit up your project. I've found the best workflow for agents is this:
1. Create detailed prompts or chains of prompts that call
2. Be sure that prompt calls targeted microtools that fetch info or do something and return it to Claude
3. Chain it to other prompts or outputs.
It uses MCP servers but these are often too open-ended. As soon as you let Claude start figuring out the strategy and the tactics you get into trouble real fast so I only have:
1. context7
2. sequential thinking
3. playwright
It's all still Alpha but I am using it to build Flashbacker every day and several other projects much much faster and cleaner. 🚀
AgentSea retweeted
Scaling AI research agents is key to tackling some of the toughest challenges in the field. But what's required to scale effectively? It turns out that simply throwing more compute at the problem isn't enough.
We break down an agent into four fundamental components that shape its behavior, regardless of specific design or implementation choices:
- Environment: The context (infrastructure) in which the agent operates
- Search Policy: How the agent allocates resources
- Operator Set and Policy: The available actions the agent can take and how it chooses among them
- Evaluation Mechanism: How the agent determines whether a particular direction is promising
We specifically focus on ML research agents tasked with real-world machine learning challenges from Kaggle competitions (MLE-bench). What we found is that factors like the environment, the agents’ core capabilities (the operator set), and overfitting emerge as critical bottlenecks long before computational limitations come into play.
Here are our key insights:
🔹Environment: Agents can't scale without a robust environment that offers flexible and efficient access to computational resources. For instance, simply running the baseline agents in the (open-sourced) AIRA-dojo environment boosts performance by 10% absolute (30% relative)—highlighting just how crucial the environment is.
🔹Agent design and core capabilities: Resource allocation optimization only matters if agents can actually make good use of those resources. Our analysis shows that the agents’ operator set—the core actions they perform—can limit performance gains from more advanced search methods like evolutionary search and MCTS. We achieve SoTA performance by designing an improved operator set that better manages context and encourages exploration, and coupling it with the search policies.
🔹Evaluation: Accurate evaluation of the solution space is critical and reveals a significant challenge: overfitting. Ironically, agents that are highly effective at optimizing perceived values tend to be more vulnerable to overfitting—a problem that intensifies with increased compute resources. We observe up to 13% performance loss due to suboptimal selection of final solutions caused by this issue.
🔹Compute: Providing agents with sufficient compute resources is essential to avoid introducing an additional limitation and bias into evaluations. We demonstrate this through experiments in which we scale the runtime from 24 to 120 hours.
In summary, successfully scaling AI research agents requires careful attention to these foundational aspects. Ignoring them risks turning scaling efforts into, at best, exercises in overfitting.
These insights set the stage for exciting developments ahead!
AgentSea retweeted
today's @wsj deciding to cover my latest essay on agency is crazy meta. it's proof that the bar to make something that can emerge - the idea of merit through action using AI, in this case - is not beyond reach. it doesn't _require_ you have stamps and credentials. i have a boring bachelor's degree in econ from a random public school in italy, and i get to work in the hottest market and on the hottest tech of the last three decades. i genuinely believe emerging today is primarily about having the willingness to do it. to be hungry, i guess
low key im so proud to see my stupid little blog grow up...
(also, so refreshing to read a pro tech take in mainstream media)
This is amazing and much needed in the world.
my weekend project to learn about bluetooth mesh networks, relays and store and forward models, message encryption models, and a few other things.
bitchat: bluetooth mesh chat...IRC vibes.
TestFlight: testflight.apple.com/join/Qw…
GitHub: github.com/jackjackbits/bitc…
AgentSea retweeted
The more I study neuroscience, the less I’m convinced that the human brain is meaningfully different than an LLM.
Recent studies show that our brain often decides to take an action before we’re even conscious of the decision.
And then we come up with rationale afterwards to explain why we did it - exactly like LLMs do 😂
E.g. if you stimulate someone’s brain to do something (e.g. lift a finger), they’ll do it and then tell you their nose felt itchy.
Excerpt from “Gray Matters” by Theodore Schwartz ⬇️
AgentSea retweeted
Eric and the team at @genspark_ai just launched AI Docs, completing their suite with AI Slides and Sheets.
It's similar to the Gemini integration in Google Docs except with a much better UX, where the AI acts more like a creative partner than just a generative tool: you get to iterate together on the output instead of just prompting once and editing the result. And it has markdown support!
If you're tired of dealing with a hundred tabs and apps for research and writing and you want to switch to vibe working instead, you should definitely check it out.
AgentSea retweeted
After AlphaGo humbled Lee Sedol, thousands of professionals pored over the alien moves; studies of 749,000-plus expert games show a sustained spike in both accuracy and creativity since 2017.
The AI that beats you today may still become tomorrow’s teacher.
It's not just machines get smarter and we stay dumb. Intelligence begets more intelligence in a virtuous loop.
pnas.org/doi/10.1073/pnas.22…
