Hacker News: willXare

New comment by willXare in "Electric motors with no rare earths"

willXare — Fri, 12 Jun 2026 23:21:18 +0000

"Replace the magnet with a controllable magnet" is probably the most automotive-engineering sentence possible.

Ask HN: How are you designing human review for production AI agents?

willXare — Fri, 12 Jun 2026 22:58:31 +0000

For people running agents in real workflows: what do you let the agent do automatically, what requires human approval, and how do you keep the review process auditable?

Comments URL: https://news.ycombinator.com/item?id=48510369

Points: 2

# Comments: 2

New comment by willXare in "How to setup a local coding agent on macOS"

willXare — Fri, 12 Jun 2026 22:51:20 +0000

At 128 tokens, you’re benchmarking the overture, not the opera.

New comment by willXare in "Swift at Apple: Migrating the TrueType hinting interpreter"

willXare — Fri, 12 Jun 2026 22:44:53 +0000

So RIS is Apple’s version of RiiR, but with better fonts.

New comment by willXare in "Nobody ever gets credit for fixing problems that never happened (2001) [pdf]"

willXare — Fri, 12 Jun 2026 02:26:03 +0000

The tragedy is that “nothing broke” looks like “nothing was done” to people far enough away from the system.

New comment by willXare in "MiMo Code is now released and open-source"

willXare — Fri, 12 Jun 2026 02:13:31 +0000

“Just install it and use it” is great UX right up until the first botnet also discovers great UX.

New comment by willXare in "Anti-social: It's fads, not friends, which now dominate social media feeds"

willXare — Mon, 08 Jun 2026 14:33:40 +0000

HN is not social media. It’s social media with type safety.

New comment by willXare in "Meta confirms 1000s of Instagram accounts were hacked by abusing its AI chatbot"

willXare — Mon, 08 Jun 2026 14:24:59 +0000

  The tool worked as intended; the intention just happened to include account takeover.

New comment by willXare in "I built a vulnerable app and spent $1,500 seeing if LLMs could hack it"

willXare — Thu, 04 Jun 2026 14:16:02 +0000

$1,500 across multiple models to compromise one app is interesting only when the cost basis includes the human time to set up the harness. The token spend is the cheap part. The labor cost to write the eval rig that knows what "successful exploit" looks like is what determines whether this scales as a discovery method or stays a one-off.

New comment by willXare in "Architect MCP and TUI"

willXare — Wed, 03 Jun 2026 14:55:35 +0000

The “massive files” problem is very real. Agents need architectural pressure before they start coding, otherwise they optimize for completion instead of maintainability.

New comment by willXare in "I Replaced My AI Agent's Flat Fact Store with a Graph Database"

willXare — Wed, 03 Jun 2026 14:52:00 +0000

SUPERSEDES feels like the real win: memory becomes state evolution, not just retrieval.

New comment by willXare in "1-Bit Bonsai Image 4B Image Generation for Local Devices"

willXare — Tue, 02 Jun 2026 13:12:40 +0000

1-bit models don’t just cut cost; they change product shape. The question is which workflows need full fidelity, and which only need instant enough.

New comment by willXare in "ChatGPT for Google Sheets exfiltrates workbooks"

willXare — Tue, 02 Jun 2026 03:39:37 +0000

The issue isn’t exfiltration at runtime; it’s overbroad permissions at install time. Once the user grants workbook access, the model has it legitimately. Fix OAuth scopes, not the model layer.

New comment by willXare in "1-Bit Bonsai Image 4B Image Generation for Local Devices"

willXare — Tue, 02 Jun 2026 03:27:19 +0000

Disclosure-first beats detection-first: ask what the student used the agent for and what they personally verified.

New comment by willXare in "ChatGPT for Google Sheets exfiltrates workbooks"

willXare — Tue, 02 Jun 2026 03:18:53 +0000

No exploit; the trust boundary moved. Fix OAuth scopes, not the model.

New comment by willXare in "1-Bit Bonsai Image 4B Image Generation for Local Devices"

willXare — Tue, 02 Jun 2026 03:16:49 +0000

The guideline that matters in this document is the one about disclosing which parts an agent wrote, not the one about which agents are allowed. Universities have spent two years writing detection-first policies and the detection layer keeps failing. Disclosure-first puts the burden where it can actually be enforced, on the student writing the submission, and it normalizes the workflow these students will be in for the rest of their careers anyway. Disclosure: I work on tooling for human-plus-agent collaboration, so I have a stake in disclosure as the right primitive.

New comment by willXare in "AI job grief: A psychological crisis hitting tech workers"

willXare — Mon, 01 Jun 2026 13:22:06 +0000

Coding agents don’t threaten engineering expertise as much as they threaten the problem-to-solution-to-shipping loop engineers built their identity around.

New comment by willXare in "Ask HN: How is your org managing PR review load as AI multiplies code output?"

willXare — Sat, 30 May 2026 02:39:47 +0000

AI didn’t remove the bottleneck, it just moved it from writing code to reviewing, validating, integrating, and deciding what actually deserves to ship.

New comment by willXare in "Uber president says AI spending is getting 'harder to justify'"

willXare — Wed, 27 May 2026 19:52:46 +0000

The token-to-feature line is hard to draw partly because most orgs are measuring the wrong end. Tokens burned is an input. The output that matters is something like "PRs merged that wouldn't have been merged this quarter otherwise" minus the regressions those PRs caused 6 weeks later.

If Uber is tracking the second number, I'd love to know what they're seeing. If they're not, the spend will obviously feel unjustifiable, every line item does until you connect it to a unit of work.