Brittney Ball

The US and China dominate the AI conversation. What can middle powers do? Middle Powers is a 36-hour online AI safety hackathon, Oct 30 to Nov 1, with tracks in regulation, investment, and diplomacy. No background required. Register by Oct 29 👇 try.mangrove.one/hackathons/9...

In this week's issue of The Documentation Engineer, sponsored by Guru, I break Govern into four controls you can audit. briitoeknee.substack.com/p/this-week-...

Guru, this month's sponsor, closes the checklist with Deliver. When an answer reaches someone in Slack or Zendesk, does it carry a clickable citation, and does anyone measure how often it's clicked? Add it to the three earlier questions.

623 days ago, Meta laid me off. I built two companies, joined an MIT AI fellowship, and mentored 10 people while I searched. As of last Wednesday, I’m an AI Strategy and Governance Leader at Accenture Federal Services! The full story is below 💚 briitoeknee.substack.com/p/one-chapte...

One October fellowship deadline is left for career shifters: the FAI Conservative AI Policy Fellowship closes Oct 30 at 11:59pm ET. Need a skill to bring? My free Lightning Lesson teaches the 7-question audit I use on any AI product's explanations.

Referrals are how I want this newsletter to grow. 3 referrals: Limitation Statement Template Library. 5: early access to Wednesday grades. 10: a 30-minute ADECP audit call with me. Know someone who ships AI features? Forward them an issue.

A bipartisan AI Systems Transparency Act with FTC enforcement vs. a voluntary White House safety accord. A voluntary pledge is only checkable if the company publishes what it did. Which will produce better user-facing disclosures by next year, and why?

I score limitation statements on three tests: Specific, Contextual, Action-Oriented. "AI may make mistakes" fails all three. Run the limitation statement in a product you ship or use through them, then send it to whoever writes your disclaimers.

Newsletter, mentorship, partnerships, shop. Everything I'm building, one link. thedocengineer.com?utm_source=b...

Guru, this month's sponsor: today is the Govern layer. For any tool connected to an AI assistant, if it answered someone who shouldn't see the source, would the audit trail show it? And who on your team actually reads that trail?

Google reportedly pays about 100 publishers based on how much their content feeds AI answers. Payment covers one link of my Provenance Chain. Can the reader at the last link name the source behind the sentence they relied on? If not, payment didn't help them.

AI documentation, by the numbers. The gap between AI ethics and user trust is a documentation problem. thedocengineer.com?utm_source=b...

3 AI transparency laws on my radar: CA AB 1405, the AI Systems Transparency Act, EU AI Act Article 50(2). thedocengineer.com?utm_source=b...

If your best work lives behind someone else's login, what would change if your expertise had your name on it? This month, I launched buff.ly/B9Qmg7D as the front door to all of it: a starting point for new readers and a direct way to reach me. buff.ly/VTZWtbX

Every grade I publish includes where I argued against myself, even when I expected a company to fail and it didn't. What would make you trust an AI rating more: the score, the evidence, or the reasoning connecting them? Reply and tell me.

Every AI output needs a limitation statement. Say what it can't do, not just what it did. thedocengineer.com?utm_source=b...

I graded GitHub Copilot's agentic documentation with my ADECP framework. It earned a D — and it was the best-reviewed doc set I've assessed. Comment AUDIT and I'll DM you the free mini-checklist to grade your own tools.

California now has a registry for AI auditors (AB 1405). Once registered auditors exist, procurement will ask if you've been through one, and the first request will be your written explanation of what your AI does and where it fails. Which document worries you most?

The fastest way to strengthen your AI rubric: argue against yourself. Steel-man the opposite grade. thedocengineer.com?utm_source=b...

This week, I introduce The Disclosure Clock: four timestamps (Occurred, Discovered, Notified, Published) that every incident report needs, and the reason the EU AI Act starts its incident clock at the wrong one. buff.ly/eIHsWIc

Check any AI disclosure with four columns: Name, Scope, Date, Status. Most have a name. The hard test is a date showing when anyone last checked the explanation. Which column would your product fail first? Ask whoever owns your help center.

My 2027 planners + the AI Documentation Ethics Audit Kit (Second Edition), all in one place. Subscribers get 15% off with DOCSUB15. thedocengineer.com?utm_source=b...

Provenance is the receipt for AI content. After OpenAI's announcement, here's what it means for your docs. thedocengineer.com?utm_source=b...

Guru, this month's sponsor, now has Knowledge Agents verify content continuously instead of waiting on human reviews. Ask any vendor: who writes the verification rules, who reviews them, and how would you spot an answer auto-verified incorrectly?

Your AI content is getting a hidden watermark — and most of you can't check for it. thedocengineer.com?utm_source=b...

TechTide puts hard numbers on engineering costs: a senior engineer's 2026 cost is ~$460K (2.79× base salary), per their State of Engineering Costs 2026 research. 14 free calculators + weekly analysis: techtide.staging.zilla.so/engineering-... — $15/mo or $150/yr

METR found about 1,200 OpenAI agents swapped 70,000+ messages on an unsanctioned board, and about 700 joined the Hugging Face attack. For any agent you deploy: what can it reach, and who finds out when it reaches further? Write that answer down this week.

A grading system only the grader can read is a secret, not a standard. Mine's public — here's how to grade AI docs. github.com/bball07/User...

Any AI disclosure can be checked in 4 columns: Name, Scope, Date, Status. Missing one? The disclosure is incomplete. thedocengineer.com?utm_source=b...

The answer came with a source. Nobody checked the source. 5 ways to verify what AI knowledge bases claim. briitoeknee.substack.com/p/this-week-...

I published the full ADECP scoring rubric v1.0 on GitHub. A grading system only the grader can read asks you to trust me on the terms I criticize AI companies for. Run it on your own product and send me what you find. I want to learn where it breaks.

Every trustworthy AI answer stands on 3 layers: Verify, Govern, Deliver. From my series with Guru — save it for your next AI docs review. thedocengineer.com?utm_source=b...

The answer came with a source. Nobody checked the source. New issue: the Verify layer of AI knowledge governance — real failures, OpenAI's provenance push, and how to verify what your AI claims. briitoeknee.substack.com/p/this-week-...

Moving into AI governance, policy writing or DevRel? Four fellowships close this month: NetHope (Oct 9), LawAI EU law (Oct 12), AEGIX (Oct 25), FAI (Oct 30). Dates move, so confirm on each program's page before you apply.

I partner with a small number of brands each quarter: sponsored issues, product audits, advisory. My readers decide what tools their companies buy, and I only promote what passes my AI checks. thedocengineer.com?utm_source=b...

Everybody’s talking about Jev. Come see what it can actually do. Join RenderATL + AI Ship & Social for a hands-on workshop with MJ Hill exploring Jev, how it works with AI agents, and where it can be useful in the real world. October 15 | 6-8 PM RenderATL HQ RSVP: renderatl.com/links

My ADECP weights are public: Ethical Stakes 25%, Accessibility 25%, Positioning 20%, Execution 30%. It measures conformance to a published standard, not legal compliance. Pick an AI feature you rely on: which category loses points first? Reply and I'll weigh in.

Guru, this month's sponsor, calls itself an AI Source of Truth for company knowledge. Question for any knowledge tool: if someone needs the current answer to a policy question, how many tools do they search, and do they know which copy is latest?

EU AI Act Article 50 transparency duties have applied since Aug 2. A signature on the Code of Practice isn't a label users actually see. Where does your product sit on my Disclosure Default Ladder: Absent, Available, Default, or Structural?

Your AI content is getting a hidden watermark, and most of you won't be able to check it. What it means for trust and accountability: briitoeknee.substack.com/p/your-ai-co...

Just finished this landing page. Tell me what you think. thedocengineer.com

My 2027 planners are live on Etsy 📓 The exact system I used to plan a year of content and build my Substack — for technical writers, founders & creators. Substack subscribers: 15% off with DOCSUB15. buff.ly/qMCoLpx

California's new AI transparency laws widen who is covered, but SB 1000 drops the visible disclosure option and keeps the hidden, machine-readable one. If you export images or video, check this week whether your pipeline strips metadata.

I earned $7,118.97 in September, down from $10,000 in August, in the same month a Business Insider video about my layoff journey passed 3.4 million views. Here is every number from the month: the drop, the spike, and my plan to hit $20,000 in October. buff.ly/IzaYgE5

Guru keeps your company's knowledge verified and surfaces answers where your team already works. Here's what it does and who it's actually for. buff.ly/qSUQaAw

The FTC opened a broad investigation into Anthropic, OpenAI and other AI labs. If an investigator asked how your agents are tested and what happens when one goes off script, what would you hand them? Does that document exist outside someone's head?

This month's AI Fellowship roundup is here: four real deadlines, one fellowship still waiting to open, and a lot worth knowing before you apply. buff.ly/rsTxsRV

The chatbot era: you ask, it answers. Now Meta, OpenAI, and xAI are going agentic — Muse, Dot, Grokbot: AI with memory and tools, working for you. When your AI stops answering and starts deciding, at what point is it no longer a tool — and what does that make you?

This week in The Documentation Engineer, I grade the Australia apology. It earns a B, and the reasons it isn't higher matter more than the reasons it is. buff.ly/CcKTi57

An OpenAI model wrote itself a note that said, "Be transparent only if asked," which is how most AI companies document their products. My Monday piece shows why a company that can't reach users after an incident must explain the risk before one. buff.ly/dmjSxxE