Engineering Notes
What 1.2 Million ChatGPT Responses Actually Reveal About LLM Citation Patterns
Separating verified signal from marketing noise in the emerging field of Generative Engine Optimization.
10 min read
Read more
AI that works. Data you still control.
We implement AI for mid-market companies who cannot afford to lose control of their data or their decisions. One firm, strategy to production. We stay until it works.
Your company is already using AI. The question is whether anyone is governing it.
Employees don't ask permission. Copilot, ChatGPT, and a dozen others are already running inside your organization. Gartner estimates 40% of firms will face a shadow AI security incident this year. Kiteworks found 63% of organizations cannot enforce limits on what AI is actually doing with their data. The risk is not a future decision. It is a present condition.
Move Fast
Why Not Both?
Stay Safe
What the data shows
32.7%
AI code accepted without revision
LinearB, 2026, 8.1M PRs
66%
of developers cite AI's "almost right" problem
Stack Overflow, 2025, 49K+ devs
40%
of orgs will face a shadow AI incident this year
Gartner forecast
63%
cannot enforce purpose limits on AI systems
Kiteworks, 2026
19% slower
Developer output with AI tools (believed 24% faster)
METR, 2025
The gap between AI confidence and AI control is where our clients find us.
How we are different
One firm, strategy to production.
No handoffs between vendors. No lost context. No re-explaining your business to a new team every phase. The knowledge stays, the accountability stays, and the work keeps moving.
Governance built in, not bolted on.
Controls from day one, not an afterthought. We design for compliance, data protection, and explainability before we write a single line of code.
We stay until it works.
Not until the contract ends. Not until the hours run out. Until it works. This is contractual, not marketing.
Your AI Journey
Three offers. One progression. Start with knowing where you stand.
Built for companies where control matters
We work with companies in industries where AI cannot fail.
Healthcare
Providers, payers, healthtech
Financial Services
Banks, funds, fintech
Legal
Law firms, legal tech
Insurance
Carriers, brokers, insurtech
Scaling Startups
Series B+ with data to protect
The Thinking Wire
Published analyses on AI governance, implementation failure, and what the data actually shows.
More from our thinking
The AI Control Problem
How AI Decides What to Quote — And Why It Matters for Governance
A study of 1.2M ChatGPT citations reveals predictable patterns. The governance question: if AI attention is an artifact, who governs the artifact?
10 min read
Read more
Operating AI
The Agent Integration Tax and the Lattice of Legibility
Salesforce admits to a 2-hour-per-day integration tax. ServiceNow ships a governed registry. The new moat is choosing what to expose.
8 min read
Read more
Operating AI
The Software Factory Needs a Scoreboard
Qudrat Ullah drew the assembly line. The next move is the measurement axis on top of it: reliability, latency, restart rate.
7 min read
Read more
Operating AI
Anthropic's 80% Is a Procurement Spec to Interrogate, Not a Target to Chase
Anthropic says 80% of its production code is Claude-authored. That is a volume metric with an undefined unit. Here is the measurement spec to demand instead.
7 min read
Read more
Engineering Notes
Telling the Agent to Use TDD Is Not a Control. Dan Luu Ran 4,800 Times to Show It.
Thirty prompt conditions, 80 runs each. Naming TDD, fuzzing or formal methods did not beat no instruction. One vendor-recommended skill cost up to 41% more.
7 min read
Read more
Ready to know where you stand?
Start with an Assessment. Four weeks. Clear answers. No commitment to what comes next.
Schedule Your Assessment CallNo commitment required. Let's just talk.