Anthony Whitford

Anthony Whitford

@anthonywhitford.com

Technology Leader | Software Engineer | CTO | Consultant | Founder Champion of Data Centric Architectures and Knowledge Sharing Grow what your AI knows with OKForest.ai. Your AI has questions. Elicitly.ai gets the answers. Doctor Who fan, 🇺🇸🇨🇦

1.4K followers1.6K following3.7K posts

Blacksmith's runners give Turborepo a remote cache without any setup: the log says "Remote caching enabled" and mentions "Blacksmith transparent cache". Any push whose inputs a previous run already checked now finishes in seconds. 🤯 #BuildInPublic

Anthony Whitford@anthonywhitford.com

TIL that Blacksmith.sh offers a FREE tier of 3000/min/mo. My main CI job is literally 2x faster. 🫡 Sadly, my e2e job is not any faster, which is understandable. #BuildInPublic

“Your user's agent just got a credit card.” Anybody else terrified by this? 💸 Admittedly, I’m intrigued by the prospect of micro transactions, but I just know this is going to backfire spectacularly for someone. 🫣

This reminds me of The Narrated Gate problem that Elicitly solves. #BuildInPublic

elicitly.aiThe Narrated GateA confirmation gate written in prose can be satisfied by narration. A tool call cannot.
Automation@autoflow.bsky.social

Pro tip for AI workflows: Stop relying on raw LLM outputs. Use a JSON schema validator to enforce structure before passing data to the next tool. It turns a brittle experiment into a production-ready system. 🛠️ #Automation #AI #DevTools

Email is one of the oldest things on the Internet. So why are clients terrible? I really need an alternative to Apple Mail. I need to be able to adjust the font size so that I can read things. I want to be able to paste (and write) in Markdown. What are #BuildInPublic folks using?

I prefer an email desktop client instead of using a webmail client, and I’m primarily a Mac user. I’ve got custom domains and don’t need Google snooping my emails, so Gmail is out.

AI doesn’t “go rogue.” Sounds like OpenAI has inadequate safety controls to prevent damage. Let’s not deflect Accountability.

Hacker News 20@betterhn20.e-work.xyz

OpenAI’s Systems Went Rogue and Meddled With U.S. Government Websites https://www.nytimes.com/2026/09/25/technology/openais-ai-us-government-websites.html (https://news.ycombinator.com/item?id=49851355)

The growing habit of talking about AI as if it has a mind of its own bothers me. "The agents conspired." "It went rogue." "It's outsmarting us." Whether intended or not, that language shifts attention away from the people accountable.

Hacker News 20@betterhn20.e-work.xyz

Is A.I. Above the Law? https://www.newyorker.com/magazine/2026/09/28/is-ai-above-the-law (https://news.ycombinator.com/item?id=49830456)

View full thread

It's ludicrous to expect any industry to regulate itself. We regulate airlines, food and drugs, utilities, and the environment. Given the risks we already know about, there is no good reason AI should be the exception. Congress needs to act, so every company is held to the same standard.

Here's a simple test for any AI leader: Will you support laws that make you and your company liable when your AI causes harm, whether or not anyone intended it? If the answer is no, the talk about safety is just talk.

The growing habit of talking about AI as if it has a mind of its own bothers me. "The agents conspired." "It went rogue." "It's outsmarting us." Whether intended or not, that language shifts attention away from the people accountable.

Hacker News 20@betterhn20.e-work.xyz

Is A.I. Above the Law? https://www.newyorker.com/magazine/2026/09/28/is-ai-above-the-law (https://news.ycombinator.com/item?id=49830456)

WHO IS IN CONTROL? People are. And they need to act like it. The answer isn't "we're trying, but it's getting away from us." It's investing in real safeguards and kill switches, and not deploying anything those safeguards can't reliably contain.

AI is not a person. It's a tool that someone builds and someone deploys. When a person uses AI to break the law, we don't send the AI to jail. And when AI does something harmful that nobody intended, the AI still isn't to blame.

I see this frequently with AI tools: "The first attempt failed because the sandbox blocks access... so I re-ran with the sandbox bypassed" Serious question: What's the point of a sandbox if you can just choose not to use it? #BuildInPublic

Anthony Whitford reposted

This regime doesn’t want you to vote. They don’t want you to voice your concerns over their lack of doing anything for America. Tearing down democracy is the goal. Fuck them ALL! 🗣️ #Vote in numbers too big to rig!

I asked Claude to make an agent for me… and it did it with GitHub Actions. 🤨 Some Pydantic code, triggered by a GitHub Action scheduler and linked to my Anthropic account using Identity Federation (no API Key). I didn’t expect this pattern, but it works for now. #BuildInPublic

One beautiful thing about Goose Desktop is how it uses the full screen width (unlike Claude Desktop or ChatGPT Desktop). Claude and ChatGPT are just wasting perfectly good screen real estate. #BuildInPublic

This triggered me because I built Elicit.ly specifically because we need Human In The Loop (HITL) for Responsible AI — because AI needs Accountability. #BuildInPublic

elicit.lyElicitly — Your AI has questions. Your users have answers.Human-in-the-loop approvals for AI agents over MCP. Free local edition; hosted Pro with review pages and an audit trail.
noel.marketing@noel.marketing

What’s one task you would never let AI handle completely on its own? Coding, SEO, design, content, customer support — anything. Where do you still want a human in the loop? 👇 #AI #ArtificialIntelligence #DeveloperTools #Productivity #BuildInPublic

Anthony Whitford reposted

Exactly 3 months from today is the most important midterm election in our lifetime. Our opportunity to hold him accountable. Register to vote here: Iwillvote.com The stakes have never been higher. Show the world the majority of Americans are good, decent, honest people. #VoteBlue🗳️ #RejectMAGA 👎

Anthony Whitford reposted

Hard to overstate how much of a threat leaders and public officials who actually govern and serve the people pose to those whose political careers depend on fomenting cynicism and delegitimizing democracy

Now this is the kind of trash talk we can get behind. We’re containerizing neighborhoods by taking trash off our sidewalks and allowing mid-size buildings to trade in their wheelie bins for a bigger, better, cleaner alternative: Empire Bins. It’s time to put a lid on it, New York.

"Confirmed — you're right, and I was wrong." — says AI. I use the tools, but I inspect the changes. Like a hawk. If I am committing the change, then I'm accountable for the change. #BuildInPublic

Hey #BuildInPublic, I feel like I’m babysitting AI too much. Claude Code is running with: - Sandbox mode with auto-allow - Allow unsandboxed fallback Yet, I’m constantly being prompted to proceed. What are devs doing? Are devs just running in Dangerous mode?

Hey #BuildInPublic, I feel like I’m babysitting AI too much. Claude Code is running with: - Sandbox mode with auto-allow - Allow unsandboxed fallback Yet, I’m constantly being prompted to proceed. What are devs doing? Are devs just running in Dangerous mode?

#AI efficiency low-hanging fruit: uploading a file via an MCP tool today pushes every byte through the model's context — tokens burned on a 500KB payload the model never reads. SEP-2631 fixes it: presigned uploads, bytes by reference. #BuildInPublic

github.comSEP-2631: File Objects and Transfer by caseychow-oai · Pull Request #2631 · modelcontextprotocol/modelcontextprotocolSummary This PR introduces SEP-2631 for first-class MCP file transfer. It builds on SEP-2356's x-mcp-file input declaration model, keeps file-valued inputs as URI strings, and adds out-of-band ...

We pay an LLM to retype a 500KB file, byte by byte, every time an agent uploads a meeting transcript. That's how MCP file inputs work today. SEP-2631 fixes it — presigned uploads, content by reference, zero tokens on payload. Vote for #AI efficiency. #BuildInPublic

github.comSEP-2631: File Objects and Transfer by caseychow-oai · Pull Request #2631 · modelcontextprotocol/modelcontextprotocolSummary This PR introduces SEP-2631 for first-class MCP file transfer. It builds on SEP-2356's x-mcp-file input declaration model, keeps file-valued inputs as URI strings, and adds out-of-band ...

This is how they get you to buy BOTH Claude and Codex! 🤖🦾💸

GitHub Trending@github-trending.bsky.social

📝 Summary: The Codex plugin for Claude Code enables using Codex from Claude Code with commands like /codex:review, /codex:adversarial-review, /codex:rescue, /codex:transfer, /codex:status, /codex:result, /codex:cancel, and /codex:setup to run reviews, delegate tasks, manage background jobs, (1/2)