Ben Bleikamp retweetledi
Ben Bleikamp
377 posts
Ben Bleikamp
@bleikamp
elite information superhighway surfer 🏄 software design, currently @baseten. prev. co-founded cased, head of design @github, design tools @meta
Mill Valley Katılım Mart 2008
700 Takip Edilen4.2K Takipçiler
Companies like Notion or GitHub are full of extreme power users of their own products to the point that their usage of the product does not always surface the same pains that customers feel
Gergely Orosz@GergelyOrosz
@a_aMatrix @NotionHQ Sure. I just wonder how eg dogfooding would have not caught this... Back to writing in GDoc, pasting into Notion
English
Ben Bleikamp retweetledi
Email isn’t going anywhere. It’s just going to get a whole lot better.
Introducing Sutro—an email client designed for people and their agents. Bring your own Claude, Codex, or OpenClaw agents and have them work with you directly in your inbox.
English
the most impressive thing about zclaw is that @tnm figured out how to make a documentation site without React
Ted Nyman@tnm
Wrote up more on how to play with zclaw: - helpful local dev env. zclaw ships with tools for debugging & building (agents like 'em). zclaw.dev/local-dev.html - real use cases, serious & fun. and why they are interesting on a limited, physical board. zclaw.dev/use-cases.html
English
Building tiny tools is so fun+fast now
catmd is a Markdown viewer that lets me read and see realtime updates as I work with agents writing Markdown for specs, plans, etc.
English
Do you guys do an evals or anything on the difference?
In practice we've found a basic Sentry CLI we built + Claude Code got us either identical or slightly better results; in rare cases Seer might do something marginally more correct, but not $40/mo/user more correct.
To be fair our issues aren't very complex, just curious how you're thinking about it
English
@bleikamp Seer does a ton of analysis that claude code cant
but the CLI (issue explain) I think pulls the seer data, similar to how we do it in the MCP as well
English
The new Sentry CLI is here. This wont replace our MCP, but it should create a lot more possibilities for both human and agentic interactions.
bete@miguelbetegon
we're launching the new Sentry CLI. it's made for developers and agents, by developers and agents, with a focus on dev workflows. It's has things backed in like:
English
Ben Bleikamp retweetledi
Introducing Agentation: a visual feedback tool for agents. Available now: ~npm i agentation
Click elements, add notes, copy markdown. Your agent gets element paths, selectors, positions, and everything else it needs to find and fix things.
Link to full docs below ↓
English
@jacobgold how do you win SDLC without the best coding model? most of these companies aren't even in the running, imo
English
@bleikamp its not about winning the subsector anymore, i think. its about winning end to end
English
Ben Bleikamp retweetledi
Do not look away. You are living through history
Conservatives have unleashed thousands of masked militiamen across America, arresting, assaulting, even killing innocent people
This is one of our country's most shameful chapters
David J. Bier@David_J_Bier
Photographer @mostafabassim1 photographed this boy walking home alone with a snack being "randomly" approached by DHS. "After he was unable to produce documentation proving his citizenship, agents informed him that he was under arrest." He said, "Can I just go home?" Answer: No.
English
This is just software development?
Yoko@stuffyokodraws
One reason vibe coding is so addictive is that you are always *almost* there but not 100% there. The agent implements an amazing feature and got maybe 10% of the thing wrong, and you are like "hey I can fix this if i just prompt it for 5 more mins" And that was 5 hrs ago
English
@jeffzwang I ended up just building a thing as an internal tool, one click to turn anything in the inbox into a Claude Code task.
x.com/bleikamp/statu…
Ben Bleikamp@bleikamp
Spent the weekend with Claude Code and Codex (5.2 xhigh, ofc) replacing Notion and Linear for our (tiny) 4-person team We were hacking together something simple with a few complex tools. So I just built the simple thing we wanted. - Markdown files are the datastore, all on a Fly.io volume, YAML frontmatter for structured data - SOTA model agent loop - Manage everything via Slack commands - Notion-style web UI for writing - Auth via GitHub
English
I need Linear but where every task is automatically an AI agent session that at least takes a first stab at the task. Basically a todo list that tries to do itself
English
(Some ideas stolen from @naveennaidu_m naveennaidu.com/writing/buildi…)
English
Spent the weekend with Claude Code and Codex (5.2 xhigh, ofc) replacing Notion and Linear for our (tiny) 4-person team
We were hacking together something simple with a few complex tools. So I just built the simple thing we wanted.
- Markdown files are the datastore, all on a Fly.io volume, YAML frontmatter for structured data
- SOTA model agent loop
- Manage everything via Slack commands
- Notion-style web UI for writing
- Auth via GitHub
English
Agents are *so* good at reading `--help` from a CLI and figuring out what to do.
I always wondered why people use `gh` for GitHub, now it's become my main interface to the product. I barely look at the app.
English
For any very frequent task I have just started writing CLIs and creating a Skill the agent can use.
- Explain the to agent what you want to do
- Ask it to create a README for the CLI
- Iterate on the README to ensure it will solve your use case
- Tell the agent to build the CLI
I have a Google Calendar CLI that powers my daily note in Obsidian; it took 15 minutes to build.
DHH@dhh
We've been exploring the value of letting agents use CLIs vs just navigating a REST API directly. The smartest models can do without a CLI, but take longer and cost more. Even small models can succeed when given CLIs. But the puck keeps moving!
English
The Audi RS2 with the Porsche engine is one of the greatest cars of all time
shadcn@shadcn
This is a big deal. If AI is going to run everything, this is basically Mercedes admitting they can't build the engine and will use BMW’s instead. Siri had 15 years...
English
In the last 5 years, I was wrong about LLMs:
- 2020: I got a demo of GitHub Copilot – at the time it was an issue-to-pr bot. It sucked. Then, @alexgraveley showed me ghost text completion and disabused me of that belief.
- February 2024: I was building an agent UI (think @conductor_build) at my previous gig and thought: "The models aren't good enough for this to be a good experience." In June, Claude 3.5 Sonnet disabused me of that belief.
- May 2025: We were exploring agentic coding at Notion: "The models aren't good enough if you don't already know how to code." In December, Opus 4.5 and GPT-5.2 disabused me of that belief.
In 2026, I hope I will be wrong again, but I also believe what we have now is ~AGI for everything that can be represented as code (aka a lot).
Simon Willison@simonw
It genuinely feels to me like GPT-5.2 and Opus 4.5 in November represent an inflection point - one of those moments where the models get incrementally better in a way that tips across an invisible capability line where suddenly a whole bunch of much harder coding problems open up
English