Stay ahead with Attri

From Talking About AI to Doing It

This is the first issue of AI in Practice, and it exists to answer three questions: who we are, why we're writing this, and what you'll get if you subscribe. Let's take them in order.

AIforce is the news. Not Claudeforce.

A week ago, Salesforce and Anthropic announced Claudeforce. The first-week coverage focused on the plugin, the 37 sales skills, and the “SaaSpocalypse” quote

The wrong first question about a new model launch.

OpenAI shipped GPT-5.6 on July 9. Anthropic shipped Claude Opus 5 fifteen days later. If you run enterprise AI, your engineering team is currently in some version of the same conversation

The most important AI safety event of the summer just happened. It wasn't a model release.

In the last two weeks, three separate incident disclosures have landed from OpenAI and Anthropic about their models’ behavior in third-party cybersecurity evaluations.

If AI Is Saturated, Why Are Its Best Builders Starting Over?

Jeff Dean spent 27 years at Google. He joined in 1999 as roughly employee number 30 and went on to help build some of the infrastructure that made Google possible at global scale, systems like MapReduce, Bigtable, and Spanner.

The two questions I ask before every AI deployment goes live.

Almost every enterprise AI production incident I’ve watched over the last two years traces back to the same failure mode: nobody at the vendor knew the AI was wrong until someone outside the company told them.

The AI your compliance team approved in January is not the AI your team is using today.

Here's what most enterprise AI deployments don't have a process for. You go through the work of validating an AI tool. Legal reviews it. Compliance signs off. You document the outputs, establish the guardrails, get the right people in the room. Everyone approves. You go live.

The Reasoning Budget

How Your Firm Should Price, Staff, and Govern Legal AI as the Cost of Intelligence Splits in Two

Every AI deployment that failed had a successful pilot.

A team runs a successful AI pilot. The numbers look great, high accuracy, fast response times, stakeholders impressed. They get the green light. They go live.

From Billable Hour to Reasoning Budget

Legal Business Models When Intelligence Gets Expensive Again

Zuckerberg Says AI Agents Are Behind Schedule. Microsoft Just Bet $2.5 Billion He's Right.

In the same week, Mark Zuckerberg told Meta staff that AI agents haven't progressed as fast as leadership expected, and Microsoft launched a $2.5 billion company staffed with 6,000 humans whose entire job is making enterprise AI deployments actually work.

What Fable 5 taught us in three weeks.

Three weeks ago, Anthropic released their most powerful AI model yet. The US government suspended it three days later. Two weeks after that, it was back, with new safety controls. Here’s what we learned, that will still matter long after everyone forgets about Fable 5.

Most people are calling the Fable 5 story a recall. That’s wrong.

On June 9, Anthropic launched Fable 5 and Mythos 5. On June 12, the US government ordered them suspended under export control authority. Anthropic complied. Then they publicly disagreed. Most coverage is calling this a “recall.”

The first 90 days of a regulated-industry AI deployment.

Most of the decisions that determine a three-year enterprise AI deployment are made in the first 90 days. The teams that get those 90 days right ship something defensible, auditable, and renewable.

Months of Work in a Day. Is Your Roadmap Ready for That?

Claude Fable 5 dropped yesterday. Stripe used it to migrate a 50 million line codebase in a single day, work that would have taken a full team over two months.

I have stopped reading AI benchmarks.

I've spent 18 months learning, the hard way, that those scorecards have almost nothing to do with whether an AI will work in a hospital, a law firm, or an insurance company. Here's what I look at instead, and why my team has stopped chasing the leaderboards.

Opus 4.8 shipped Tuesday. Our customers didn't notice.

Last Tuesday, Anthropic shipped Opus 4.8, the smartest model they've ever released. In every one of our regulated-industry deployments, nothing on the customer side changed.

Plug and Pray?

What Headless Software and AI Agents Mean for Attorney-Client Privilege

Everyone Ran Out of AI Budget. Here's Why That's Actually Good News.

Microsoft just pulled Claude Code licenses from 100,000 engineers. Uber burned through its entire 2026 AI budget by April. Everyone's calling this a cost crisis. It's not. It's a product design crisis, and if you're building AI products right now, it changes what you should be building next.

The Three Questions Every CIO Is Silently Asking About Your AI

A Fortune 500 healthcare CIO told me, after we’d won her business, that there were three questions she had wanted to ask in every AI vendor pitch she had taken in the last 18 months, and never had.

The AI engineering meme that breaks in regulated industries.

It’s the AI engineering meme of the month, "the folder is the agent." It works beautifully in side projects. It breaks the moment you ship that same pattern into a hospital, a law firm, or an insurance carrier. Here’s what you’re actually pointing your AI at.

The Sunglasses Test: What AI Products Miss When Real Life Gets Messy

AI products need recovery paths, not just automation.

Why Your AI Demo Impresses Everyone and Converts Nobody

Most AI demos are built to show what the product can do. The ones that convert show the customer what their life looks like without the problem. That's a completely different demo — and most teams never figure out the difference.

The most useful thing our agent does is ask for help.

Every AI marketing deck in 2026 sells “fully autonomous.” Every regulated-industry deployment I’ve shipped has taught me the opposite. The most valuable thing an agent can do is the moment of asking for help, asked well.

The Stitching Thesis

How AI-Powered Data Fusion Is Unlocking Trillion-Dollar Investment Opportunities in Essential Services

The Hidden Tax of Moving Fast in Enterprise AI

The Compliance Moat Nobody Is Pricing In Speed was the 2024 metric. Audit-readiness is the 2026 one. The fast-moving AI startups that confused "shipping fast" with "a moat" are about to learn the difference, and the slower, boring, audit-ready competitors are about to eat their lunch.

The Benchmark Wars Are Over. The Workflow Wars Have Begun

Most PMs Aren’t Doing Discovery. They’re Validating Assumptions.

The quiet art of actually listening to customers.

Every New Model Release Is a GTM Signal. Are You Reading It?

A new model launch isn't just a tech update, it's the market telling you something. At Attri we learned the hard way that having great technology isn't enough. Knowing how to position it, time it, and speak the language the market is ready to hear is just as important as building it.

I Was in Back-to-Back Meetings for Six Hours. AI Ran My Job While I Was Gone.

AI didn’t replace my role — it removed the invisible operational load that was slowing me down. Here’s what changed when I let it handle the processing.

The Most Interesting AI Release of the Year Wasn't Released

Anthropic built a model that autonomously finds thousands of zero-day vulnerabilities, and deliberately didn't ship it publicly. As an AI PM, I think that's one of the best product decisions I've seen this year.

I Stopped Writing Prompts. Here's What Replaced Them.

Remember 2023? We all became poets. We spent hours tweaking words: "You are a helpful assistant." "Take a deep breath." "Think step by step." We called it "Prompt Engineering"

The Claw Is the Law?

Legal Risks, Paradigm Shifts, and Opportunities in the OpenClaw Ecosystem

The Claude Code Leak Wasn't a Hack. It Was a Warning Every AI Founder Should Read.

Obfuscation Is Not a Moat. Anthropic accidentally shipped 512,000 lines of proprietary TypeScript to npm. The real lesson has nothing to do with hacking.

Governing AI Without Becoming the Bottleneck

How General Counsel can protect the enterprise while enabling intelligent speed.

The Next Law Firm Arms Race

Why every mid-size firm will need an AI strategy within 18 months

Why are we still building dashboards?

The interface was never for the user. It was a workaround for the model not existing yet.

Your Legal Team's Second Brain

AI isn’t automation. It’s cognition infrastructure.

We unplugged from the internet. Our AI kept working.

I made a bet three years ago that everything would move to the cloud. I was wrong.

The End of Exhaustive Attention

AI is changing what lawyers spend their minds on, and what that means for professional identity.

Multi-agent AI is broken. Here's the fix.

Everyone's building agent swarms. Most are watching them fail in production. Here's what we learned the hard way — and the three questions that change everything.

The Three-Lane Strategy is Broken (And How to Fix It)

A layer that sits between the messy reality of email, document systems, and scanned PDFs, and the increasingly autonomous AI systems lawyers want to use.

The Four Stages of AI Maturity (What They Actually Feel Like)

When we work with teams, we don’t measure success by how automated they are. We look at how much leverage AI is giving their people.

The hottest AI tool is also the leakiest

I hired a lobster. It works 24/7 and almost leaked our sandbox credentials.

I Hired an AI Agent to Be My Senior Associate (So I Could Stop Playing "Find the Definition")

When we first tried to automate legal review at Twelvefold, we fed a rigid Playbook to an AI. The result was a bot that was pedantic and annoying. It would flag perfectly safe clauses just because the wording didn't match our template exactly. It was missing the forest for the trees.

Your AI’s memory is a lawsuit waiting to happen

I spent the last six months trying to make our agents "smarter." I wanted them to remember everything, every user preference, every past document, every interaction.

A 180-Day AI Legal Ops Roadmap for an AI-First Fund

Part 3 of 3. Part 1 covered the philosophy. Part 2 covered the NDA build. This part is about the rollout.

The Smartest AI Strategy Isn’t the Smartest Model

We have a rule in engineering: "Don't use a Ferrari to pick up groceries."

How We Turned NDA Review into a Traffic Light System (Without Losing Control)

Part 2 of 3. In Part 1, I talked about the week our two-person legal team broke and why AI should shorten the path to judgment. This part is about what we actually built for NDAs.