NSSG Insights
Operational Research.
No Strategy Decks.
Research-backed analysis on enterprise AI rollout, scaling failures, and what separates the 5% that produce results from the 95% that don't.
21 of 21 tasks done, and the three biggest never ran
A subagent self-reported 21/21 tasks complete. Grep-verification found the bulk of the work never ran. Here's what I changed after.
Read articleOne cd into a submodule bricked every tool for the session
A plain cd into a git submodule invalidated my shell's working directory and killed every tool call after it. Here's what I learned about chaining commands.
Read articleMy Agent Confirmed a Page It Never Saw
173 tests passed, two agents reported the page as visually confirmed, and the page was broken the whole time. The story of how a subagent signed off on a render nobody had actually looked at, and the boring fix that catches it every time.
Read articleA Synced Folder Corrupted My AI Coding Sessions
For months it came and went. Weeks would run clean, then a long autonomous session would throw access-denied errors that spread to my other sessions. I chased the tooling for a long time before I found the cause, which was a synced folder, and the thing it left behind was my trust in what the factory produced.
Read articleAn Agentic Coding Workflow's First Six Months: From Template to Surviving Opus 4.7
The agentic factory began as a generic template in November 2025. Six months later it runs 29 specialist agents across a 17-phase workflow and absorbed an Opus 4.7 model migration with zero factory changes. How it grew up.
Read articleOur 17-Phase AI Coding Factory Survived an Opus 4.7 Migration Without Any Changes
We switched to Opus 4.7 mid-workflow and shipped a Complex-tier feature in 12 hours with no factory changes. What held, and the two issues that surfaced.
Read articleWhy 95% of AI Pilots Fail to Scale. What Operators Do Instead
MIT research finds 95% of enterprise generative AI pilots are failing to accelerate revenue, with separate BCG and S&P Global findings pointing the same way. The cause isn't the technology, it's the operating model. Here's what separates the 5% that scale.
Read articleWhy Your Team Keeps Solving the Wrong Problem
Rework can cost as much as 2.2% of annual revenue. A large share of that rework traces to errors in the requirements phase, before any code is written. The case for problem definition as the highest-ROI activity in any project.
Read articleThe Grid Can't Keep Up: Inside AI's Data Center Energy Crisis
AI data centers consumed 415 TWh in 2024, demand is projected to double by 2030. Fossil fuels, nuclear restarts, waste heat recovery, the incentive gold rush, and why 10+ states are now imposing moratoriums on new construction.
Read articleAI in Financial Services: Where the Returns Are Real and Where They Aren't
Banks spent $31.3 billion on AI in 2024. Only 29% report meaningful cost savings. Fraud detection is delivering. Generative AI is officially in the Trough of Disillusionment. Here's what the data actually shows.
Read articleWhat Does an AI Rollout Actually Cost?
Enterprise AI costs range from $50K to $5M+ annually, but 85% of companies misestimate by more than 10%. Here's the real breakdown by company size, industry, and what's hiding in your budget.
Read articleWhy Process Improvement Projects Fail Before They Start
The famous '70% of change initiatives fail' stat has no empirical basis. The real data is worse: Bain tracked 24,000+ initiatives and found only 12% achieved their original ambitions. Here's what actually kills improvement projects.
Read articleThe Process You Don't Document Is the Process You Can't Improve
Undocumented processes cost $2,400 per process per year in lost productivity, before accounting for AI failure. Gartner forecast that 85% of AI projects would deliver erroneous outcomes through 2022, driven by bad data. The business case for documentation as operational infrastructure.
Read articleAI in Healthcare: Where the ROI Is Real and Where It Isn't
CommonSpirit saved $100M. Nebraska Medicine freed inpatient capacity without construction. But 80% of healthcare AI projects fail to scale. Here's exactly where the money gets made and where it disappears.
Read articleThe AI Governance Question Nobody Asks Until It Kills the Project
Only 12% of executives can identify the right controls for common AI risks. Air Canada paid for it in court. McDonald's paid at 100 locations. Here's what governance failure actually looks like, and how to fix it in 90 days.
Read articleHow to Measure AI ROI Before You Start
Only 29% of companies can confidently measure AI ROI. Here's the pre-implementation framework CFOs and boards actually use, KPIs, baselines, hidden costs, and what to define before you spend anything.
Read article