Hunter Harris

Vision3D with full Muse support update is now live on the App Store.

@markgurman @fotsch1 @neilcybart keep pushing your grubby little subscription

@fotsch1 @neilcybart You left Apple in 1996 while Ternus was still in college and you’re citing your time at Apple as being able to declare some knowledge on this topic? Give me a break. Enjoy your subscription.

This is BS. If you want to know the unsensationalized truth, follow @neilcybart He is BY FAR the most knowledgeable person about 🍏, outside of 🍏. They’d why people pay for his reports and no one else’s. — engr, decade @ AAPL

@virtual_amir @vrhermit @Apple @Logitech Very nice Amir! How do you sync the origin point for all users?

Made on the @apple Vision Pro with @Logitech Muse. Recorded real time on iPhone companion app (calibrated to paper on floor 😂🫣). Happy Nowruz! #calligraphy #applevisionpro

What if agents could customize the user interface?

Custom commands. Every time you catch yourself typing the same prompt twice, stop and make it a Claude slash command. Your workflow compounds. Day 1 you have /test. Day 30 you have 50 commands and you're shipping faster than teams of 5.

Labs. Get Claude to build swipeable views where you can test multiple designs side by side. Don't pick a UI in your head, generate 5+ variations, swipe through them on device, pick the one that feels right. Design decisions in seconds instead of hours. Get Claude to create a lab based approach from your idea.

Most vibecoding advice is "write better prompts." That's level 1. Here's what actually levels you up. Build infrastructure around the AI, not just with it.

Delve, a YC-backed compliance startup that raised $32 million, has been accused of systematically faking SOC 2, ISO 27001, HIPAA, and GDPR compliance reports for hundreds of clients. According to a detailed Substack investigation by DeepDelver, a leaked Google spreadsheet containing links to hundreds of confidential draft audit reports revealed that Delve generates auditor conclusions before any auditor reviews evidence, uses the same template across 99.8% of reports, and relies on Indian certification mills operating through empty US shells instead of the "US-based CPA firms" they advertise. Here's the breakdown: > 493 out of 494 leaked SOC 2 reports allegedly contain identical boilerplate text, including the same grammatical errors and nonsensical sentences, with only a company name, logo, org chart, and signature swapped in > Auditor conclusions and test procedures are reportedly pre-written in draft reports before clients even provide their company description, which would violate AICPA independence rules requiring auditors to independently design tests and form conclusions > All 259 Type II reports claim zero security incidents, zero personnel changes, zero customer terminations, and zero cyber incidents during the observation period, with identical "unable to test" conclusions across every client > Delve's "US-based auditors" are actually Accorp and Gradient, described as Indian certification mills operating through US shell entities. 99%+ of clients reportedly went through one of these two firms over the past 6 months > The platform allegedly publishes fully populated trust pages claiming vulnerability scanning, pentesting, and data recovery simulations before any compliance work has been done > Delve pre-fabricates board meeting minutes, risk assessments, security incident simulations, and employee evidence that clients can adopt with a single click, according to the author > Most "integrations" are just containers for manual screenshots with no actual API connections. The author describes the platform as a "SOC 2 template pack with a thin SaaS wrapper" > When the leak was exposed, CEO Karun Kaushik emailed clients calling the allegations "falsified claims" from an "AI-generated email" and stated no sensitive data was accessed, while the reports themselves contained private signatures and confidential architecture diagrams > Companies relying on these reports could face criminal liability under HIPAA and fines up to 4% of global revenue under GDPR for compliance violations they believed were resolved > When clients threaten to leave, Delve reportedly pairs them with an external vCISO for manual off-platform work, which the author argues proves their own platform can't deliver real compliance > Delve's sales price dropped from $15,000 to $6,000 with ISO 27001 and a penetration test thrown in when a client mentioned considering a competitor

@comet I could have built this in 72 hrs, hmu. dropping soon.

Computer now works inside Comet with full browser control.

@TheAdwaithVarma @comet Wow it looks like they built this in like 3 days. Dropping something much better soon.

Using @comet Assistant for shopping suggestions, on iOS.

Almost 1 month since PolyAce Chess launched (Feb 20) Next theme set is coming soon: NEON. A futuristic neon city + sleek night atmosphere. Play PolyAce Chess free on Apple Vision Pro while you wait! #PolyAceChess #VisionPro #AppleVisionPro #Chess

the part that still surprises me: Claude reads the screenshot. not OCR, it describes what the broken UI actually looks like and uses that as evidence. it's not a just a tool. it's a junior on-call engineer that never sleeps.

then it reconstructs the timeline. plots the network failures, iOS errors, and server logs on the same axis, filtered to ±30 minutes of when the user filed the report. ranks every signal by confidence: HIGH / MEDIUM / LOW, with an evidence chain.

Here’s how I made a Claude skill that debugs user crash reports end-to-end screenshots, production logs, and a file:line fix. one command, under a minute.

40+ skills. Autonomous SRE. Adversarial pentesting. Competitive parity tracking. Growth analysis. UX audits that walk every screen like a real user. A skill that creates other skills. But the pattern that matters isn't any individual skill. It's the loop: RUN → EVALUATE → LEARN → PERSIST → REPEAT Every skill maintains state across sessions. Every skill feeds findings back into the system. Every skill gets better the more you use it. Claude Code skills aren't prompts. They're autonomous systems with memory, self-improvement, and structured self-disagreement. Build the machine that builds the machine.

/chat-rl This is the one that changed everything. /chat-rl sends real messages to my AI assistant across 8 tool categories — web search, places, calendar, email, comparison shopping, image generation, smart home, vehicle commands. Then it grades every response on a strict rubric: A — Perfect tool use, correct result B — Right tool, minor issues C — Partially correct F — Wrong tool, hallucinated result, or timeout After grading, it submits ratings to the production RL feedback API. Score of 1.0 for good responses, -1.0 for bad ones. These ratings feed back into the experience store that shapes future agent behavior. The agent literally learns from its own test results. It maintains a BAD-IDEAS.md — an anti-pattern registry documenting every failed approach. "tool_results can be null, not just empty array." "Comparison mode oscillates between 2 and 3 search loops unpredictably — do not try to fix this, it's a feature." "60-second timeout is the #1 failure mode, not tool selection." 8 consecutive all-green sessions. 26 total sessions. Every session submits 8 ratings. That's 200+ labeled training signals fed back into the system — not from human reviewers, but from structured automated evaluation. The agent improves itself by testing itself. That's not CI/CD. That's evolution.

I have 40+ custom Claude Code skills. One trains my AI agent through reinforcement learning. One grades my browser automation across 50 scenarios. One literally keeps a Hall of Shame for when Claude embarrasses itself. My top 5 most useful — and why they changed how I build: