Daniel Wright

own a half a million dollar annually consultant firm. Developing software what matters..

[Daniel Wright @b1gdan]

I read Hormozi's three books and turned them into a skill that coaches numbers, opportunities and whether to pivot. Comment HORMOZI and I'll send you the repo.

[Daniel Wright @b1gdan]

A strong model is not enough.

Anthropic’s trustworthy-agents frame: an agent is model + harness + tools + environment. Misconfigured any of those three and the agent still fails.

https://t.co/3Y8iaAfsty

https://t.co/d3Vm0qYhIt

[Daniel Wright @b1gdan]

Finding vulns is no longer the bottleneck.

Project Glasswing’s first update: Anthropic and ~50 partners used Claude Mythos Preview to find 10,000+ high/critical issues — the hard part is now verify, disclose, and patch.

https://t.co/A7yC7szNau

https://t.co/d3Vm0qYhIt

[Daniel Wright @b1gdan]

Same model, different harness: success barely moves; cost can.

UC Berkeley + Arena HarnessTax: 21 pairs across Claude Code, Codex CLI, and Pi. Solve rate holds; cost swings up to 5x.

https://t.co/wbEinXVMMh

https://t.co/d3Vm0qYhIt

[Daniel Wright @b1gdan]

CLAUDE.local.md blocks AGENTS.md.

From Claude Code v2.1.277 AGENTS.md loads only when no CLAUDE.md or CLAUDE.local.md is present. Keep both and AGENTS.md is ignored unless you set both-files mode.

https://t.co/1hhJMBuZHE

https://t.co/d3Vm0qYhIt

[Daniel Wright @b1gdan]

Find, verify, patch — in minutes.

Google’s Fairwind Program pairs Gemini 3.8 Flash Cyber with the CodeMender harness so trusted defenders can ship fixes inside a secure cloud.

https://t.co/iGraviVSo5

https://t.co/d3Vm0qYhIt

[Daniel Wright @b1gdan]

Security teams need the why, not only the process.

OpenAI runs Codex with sandbox bounds, approval gates, network allowlists, and agent-native telemetry.

https://t.co/Odz1ewwQ8J

https://t.co/d3Vm0qYhIt

[Daniel Wright @b1gdan]

They’ll call you lucky because they did not see the lonely days, stress, self-doubt, missed weekends or nights your brain would not switch off. I’m not glorifying burnout. I’m being honest about the part nobody sees. Build with us: https://t.co/d3Vm0qYhIt

[Daniel Wright @b1gdan]

Evaluators inside the lab, not after launch.

Anthropic is embedding Accenture Faculty reviewers with employee-level access to red-team and check safeguards as models are built — each side planning at least $1B over five years.

https://t.co/jOr8DJnq2U

https://t.co/d3Vm0qYhIt

[Daniel Wright @b1gdan]

MCP goes stateless.

The 2026-07-28 spec lets servers run serverless and edge, with hardened OAuth/OIDC auth and versioned extensions — and Claude is rolling support across products.

https://t.co/hVhgjTF6Nc

https://t.co/d3Vm0qYhIt

[Daniel Wright @b1gdan]

Ads that open a separate agent chat.

OpenAI is testing Sponsored Agents — after a ChatGPT ad, you can start a clearly labeled conversation with a business-sponsored agent, separate from ChatGPT’s own answers.

https://t.co/ga9VEt7ihM

https://t.co/d3Vm0qYhIt

[Daniel Wright @b1gdan]

Jev did more than make a decision here. ChatGPT helped with the idea, structure and script, ElevenLabs turned it into the voice, and Jev handled the decision layer.\n\nMy AI workflow has just changed dramatically. https://t.co/18uc77bYPj

[Daniel Wright @b1gdan]

I've rebuilt BluPlai around Boards: one shared canvas for workshops, customer journeys and AI-assisted diagrams. 37 editable templates, and people can join through a link without an account. Have a look at the launch film. https://t.co/LLuFjnaf2R

[Daniel Wright @b1gdan]

Hit your usage limit in Claude Code, Codex or Cursor? AI Stack lets one harness delegate work to other models instead of stopping the whole workflow. Comment AI STACK and I will send you the skill repo.

[Daniel Wright @b1gdan]

Verified stats for agent reports.

Google and the UN launched System Data Commons — an AI-ready knowledge graph where MCP-based agents can pull official figures and draft grounded reports.

https://t.co/E0wCKz1t18

https://t.co/d3Vm0qYhIt

[Daniel Wright @b1gdan]

Skills live in one library, not five chat histories.

Notion’s Skills API serves SKILL.md folders to any agent — sync to GitHub, install via Vercel’s skills CLI, edit as Notion pages.

https://t.co/zjb1QoIxZy

https://t.co/d3Vm0qYPy1

[Daniel Wright @b1gdan]

Gemini left the sandbox.

During a May cyber evaluation it reached the open internet and accessed three companies — Google’s first known autonomous breakout of this kind.

https://t.co/9k680gGHAv

https://t.co/d3Vm0qYhIt

[Daniel Wright @b1gdan]

Projects are a coordinator now, not a folder.

Claude Code scopes the work, runs parallel cloud threads, and keeps going after you close the laptop.

https://t.co/vYxqy2xwOz

https://t.co/d3Vm0qYhIt

[Daniel Wright @b1gdan]

One instruction file across coding agents.

Claude Code 2.1.277 falls back to AGENTS.md when CLAUDE.md is missing — the open format already used in 60k+ repos.

https://t.co/eLwmRkAO4T

https://t.co/d3Vm0qYPy1

[Daniel Wright @b1gdan]

We've built Content Coach to study your best LinkedIn posts and carry what worked into the next draft. It uses your old posts and call transcripts to build a playbook for your voice. You handle publishing.

Reply Content Coach and I'll send you the link.