Sign in

Mike Buckbee

@buckbee.bsky.social
4.1K followers 947 following 1.7K posts

Dev+Founder+Dad -> HutchDB, Expedited Sec, Evvl Mostly post about cybersecurity and AI dev stuff.

PostsRepliesMedia
Mike Buckbee @buckbee.bsky.social · 1h
100% - it makes is significantly easier to iteratively improve the classification.
000
Mike Buckbee @buckbee.bsky.social · 1h
What are folks doing for in-repo documentation for AI building? I've bounced between Google's OKF, a docs subfolder, overloading AGENTS.MD Any Tips?
000
Mike Buckbee @buckbee.bsky.social · 05/10/2026
You can ignore the posts about Jev and whether LLM classifers are better if you remember one thing: always split your AI classification and your AI generation into separate tasks.
100
Mike Buckbee @buckbee.bsky.social · 05/10/2026
Stupid low hanging fruit. Sped up our test suite by 5x by just asking Claude 'hey can you make the test suite faster'
000
Mike Buckbee @buckbee.bsky.social · 05/10/2026
It's "Rust Token Killer" but not actually specific to rust code - github.com/rtk-ai/rtk
000
Mike Buckbee @buckbee.bsky.social · 04/10/2026
Claude, Codex, Grok, etc. are wildly inefficient running shell commands because they call the default verbose versions of them. Using RTK, I've been able to cut input tokens from them by _half_ - nevermind that this also speeds up everything as well. Recommended.
120
Mike Buckbee @buckbee.bsky.social · 03/10/2026
I'd read it.
010
Mike Buckbee @buckbee.bsky.social · 02/10/2026
Yesterday 7 of 3,727 user-facing AI API calls failed (0.19%), all upstream model timeouts. Users saw "try again later." Our p99 hit 3.7s, then recovered. Is 0.19% acceptable, or would you add a fallback model?
000
Mike Buckbee @buckbee.bsky.social · 02/10/2026
That's clever, I feel like you'd be easy to update that as well.
100
Mike Buckbee @buckbee.bsky.social · 02/10/2026
Is that PR draft shared by other devs or you + other AI agents?
110
Mike Buckbee @buckbee.bsky.social · 02/10/2026
Is that something you've put together or something provided (ala Claude's /review skill)?
100
Mike Buckbee @buckbee.bsky.social · 02/10/2026
Do you run stuff like /simplify or TDD skills or anything against the code before the PR draft?
110
Mike Buckbee @buckbee.bsky.social · 02/10/2026
Are y'all running code review + testing agents on your code before sending it to PR or waiting until after?
210
Mike Buckbee @buckbee.bsky.social · 02/10/2026
That's a good one, let me check it out.
000
Mike Buckbee @buckbee.bsky.social · 02/10/2026
Does anyone have any good examples of AI-assisted (MCP) onboarding for SaaS? Both Stripe and Posthog have gone this route but I'm trying to collect more as I think it's fascinating.
210
Reposted by Mike Buckbee
austin (eeek!/ackkk! 👻) @thebadcode.com · 30/09/2026
o11y hacks: add “number of visits to your status page” as an alert
49810
Mike Buckbee @buckbee.bsky.social · 29/09/2026
Ever suffered through the pain of asking someone to make a DNS change over email? Suffer no more! I've just added the ability to share DNS change requests on the DNSisBeautiful site. Bundles the request + a test if they set it right into one shareable URL.
011
Mike Buckbee @buckbee.bsky.social · 28/09/2026
Tell me what's bad/wrong here? I'm prototyping a new feature for DNSisBeautiful that lets folks make request links via simple parameters that they can then send to someone to get them to make a DNS change. Would this make sense if you got landed on this page?
000
Mike Buckbee @buckbee.bsky.social · 24/09/2026
Used Opus 5.5 for the past 24 hours across ~5 projects and it sure does feel like Fable smarts but much faster.
030
Mike Buckbee @buckbee.bsky.social · 23/09/2026
This is part of what makes it interesting is that you get "smart merge" on inserts instead of just insert/upsert. The AI model will respond with "What do you want to do with these records".
000
Mike Buckbee @buckbee.bsky.social · 23/09/2026
If you want to try the new Opus 5.5 with Claude Code make sure that you've installed it via the native installer and not via homebrew (which lags releases).
010
Mike Buckbee @buckbee.bsky.social · 21/09/2026
So in Claude I'd literally just say "store these to the XYZ keywords in hutch" and then it would sort out all the field names, and structure. Then I could just keep pointing new tools against the same collection. And then I could share out the final set with the client, etc
000
Mike Buckbee @buckbee.bsky.social · 21/09/2026
It's a database that's directly accessible from AI agents and tools via MCP, but that requires zero up front work on data modeling and setup. I'd built it when doing keyword research, I'd get kw data from a bunch of places and manipulate it via Claude code, but didn't have a place to store it.
611
Mike Buckbee @buckbee.bsky.social · 20/09/2026
Hey friend. Can you spare a ⭐️ for a (hopefully) interesting open source project I'm working on? github.com/ExpeditedPro...
github.com
GitHub - ExpeditedProjects/hutchdb: A database for AI agents. No schema, SQL, or structure required. Claude Code, Codex, Cursor, Hermes, Claw, Grokbot, and other MCP clients store, query, and share th...
A database for AI agents. No schema, SQL, or structure required. Claude Code, Codex, Cursor, Hermes, Claw, Grokbot, and other MCP clients store, query, and share the same records. Open source (AGPL...
171
Mike Buckbee @buckbee.bsky.social · 15/09/2026
It's kind of crazy how much better Claude Code is than Codex at handling and working with multiple different iterations of MCP servers. It's not even the "future" it's literally just what I need to get through my working day.
110
Mike Buckbee @buckbee.bsky.social · 12/09/2026
Anyone have a really good way of using worktrees in dev? I've got my own per-project setup but it feels like there's got to be a better way.
000
Mike Buckbee @buckbee.bsky.social · 08/09/2026
My new sign off on drip emails is "p.s. if you reply to this email a human will answer it." Is that actually helpful information to people signing up?
000
Mike Buckbee @buckbee.bsky.social · 08/09/2026
If you find yourself with extra AI token quota before a reset, start grinding out dependency+security updates and improving test coverage of your apps.
010
Mike Buckbee @buckbee.bsky.social · 04/09/2026
I do like some balance, so usually there's a few paragraphs in output I ignore unless the choices presented feel off.
010
Mike Buckbee @buckbee.bsky.social · 01/09/2026
Genuinely shocking to me that some people aren't using a "fast-chat" skill with ClaudeCode or Codex. Turns narrations into super quick decision points.
100
Mike Buckbee @buckbee.bsky.social · 31/08/2026
Working on HutchDB has been very frustrating as when try it out and get it they go "oh this is very handy!" but actually describing it has been a challenge. Trying a new pitch. Better?
010
Mike Buckbee @buckbee.bsky.social · 28/08/2026
One of these hops is not like the other. One of these hops blew up my support inbox today.
000
Mike Buckbee @buckbee.bsky.social · 28/08/2026
Right. Hutch is sort of a primitive to where you put all those things so that you can do things like have a *Claw or Hermes agent gather info -> put it in Hutch -> Human review -> kickoff process from GrokBot or Claude Code against it, etc.
110
Mike Buckbee @buckbee.bsky.social · 27/08/2026
Store/Fetch any arbitrary data set directly from your agent via MCP. github.com/ExpeditedPro... - please drop a ⭐️ on the repo if it sounds interesting.
github.com
GitHub - ExpeditedProjects/hutchdb: Simplest self-hostable MCP server for structured agent data. Headless, single-user. AGPL v3.
Simplest self-hostable MCP server for structured agent data. Headless, single-user. AGPL v3. - ExpeditedProjects/hutchdb
000
Mike Buckbee @buckbee.bsky.social · 27/08/2026
Most AI "Agent" demos are really slight and kind of underwhelming because they don't have access to persistent data storage and collaboration (to both human and other AI agents), which is why I'm building HutchDB.
github.com
GitHub - ExpeditedProjects/hutchdb: Simplest self-hostable MCP server for structured agent data. Headless, single-user. AGPL v3.
Simplest self-hostable MCP server for structured agent data. Headless, single-user. AGPL v3. - ExpeditedProjects/hutchdb
821
Mike Buckbee @buckbee.bsky.social · 22/08/2026
Good tips. Implementing a mix of those.
100
Mike Buckbee @buckbee.bsky.social · 21/08/2026
Dealing with bots signing up for SendCheckIt only to immediately invoke the "forgot password" form to try and email bomb people. They're also abusing the gmail dot trick to push more through. Anyone else had to deal with this?
100
Mike Buckbee @buckbee.bsky.social · 19/08/2026
I've been lucky enough to be in the Changebot beta and they were nice enough to ship my most requested feature: destination selection during publishing. updates.changebot.ai/changebot/20...
updates.changebot.ai
Setup destinations during publishing
Introducing destination setup on the Publish pageYou can now add a destination while you're in the middle of publishing and return immediately to the Publish...
000
Mike Buckbee @buckbee.bsky.social · 18/08/2026
Overall I think the codex sandboxes + permissions makes more sense over Claude's. Less friction, more freedom.
101
Mike Buckbee @buckbee.bsky.social · 16/08/2026
Codex coding users, what's the recommended reasoning level I should be on?
000
Mike Buckbee @buckbee.bsky.social · 16/08/2026
@jeffcannata.bsky.social - thought you and the filmcast would appreciate these propmasters talking about why + how the "empty coffee cup" in movies thing happens and what the fixes are after the One Night Only review -> www.youtube.com/watch?v=mlP2...
youtube.com
Explaining the Annoying Empty Cups in Movies and TV
YouTube video by Scott Prop and Roll
010
Mike Buckbee @buckbee.bsky.social · 14/08/2026
My experiments with procedural terrain generation are going really well. I'd previously produced a nice looking meadow, but now have a really slick grimdark cobblestone as well.
020
Mike Buckbee @buckbee.bsky.social · 13/08/2026
Woke up to a support request from a site that is suffering what appears to be a supply chain attack. In production, their web app is launching a second GTM script that then loads a fake captcha that's likely stealing user sessions. Check your dependencies!
120
Mike Buckbee @buckbee.bsky.social · 12/08/2026
My friend Ernesto and his Ombulabs is running a free webinar starting _right now_ on how to use Claude safely in your business - www.ombulabs.ai/webinar/safe... - you should check it out.
ombulabs.ai
Webinar: Claude for Small Business | OmbuLabs.ai
Join our free webinar to learn how non-technical teams can safely adopt Claude in their business. Get practical guidance, real-world examples, and best practices for rolling out AI responsibly.
020
Mike Buckbee @buckbee.bsky.social · 11/08/2026
We will see, for now it's mostly pipe dreams.
000
Mike Buckbee @buckbee.bsky.social · 11/08/2026
Bad news, the cobblestones now look more like cobblestones and less like mid-century modern tiling.
000
Mike Buckbee @buckbee.bsky.social · 11/08/2026
Started working on a new fun side project that involves lots of procedural image generation and hopefully ends up with launching my first Kickstarter.
340
Mike Buckbee @buckbee.bsky.social · 11/08/2026
Hit my Claude Code Max + Fable limits for the week so I'm trying Codex for the day. Any tips?
010
Mike Buckbee @buckbee.bsky.social · 10/08/2026
Added a cool little Referrer-Policy security feature to Expedited WAF this morning.
120
Mike Buckbee @buckbee.bsky.social · 07/08/2026
The real question. How many people are recursively in _his_ pocket.
000