From the desk of RoboCFO

A component-level walkthrough of how agent harnesses work in production. The action loop, the ten core components, the capability-risk spectrum, and what changes when you're building for finance.

In finance, hallucination is a verification problem. Agent harnesses, the execution infrastructure around AI models, are how finance teams solve it. Case studies from Morgan Stanley, Robinhood, and Auquan show the pattern.

Brynjolfsson, Chandar, and Chen's ADP data shows a 13 percent relative employment decline for workers aged 22 to 25 in high-exposure occupations. The succession pipeline your workforce plan assumes in 2030 is being rewritten now.

Fresh NBER data shows 69 percent of firms use AI and nine in ten executives report no realized impact over three years. The same executives forecast productivity gains ahead. A CFO's guide to budgeting through the gap.

Experimental studies time the output task. Field studies time the week. If you're budgeting AI productivity off conference-stage numbers, you're solving for the wrong work.

Accounting teams are past the "is AI ready?" phase. These ten use cases are running in production at real companies today, with measurable time savings, named vendors, and clear maturity levels. Here's what works, what's emerging, and where to start.

The AI CFO isn't a robot replacing the corner office. It's a finance leader who has restructured their operation around AI agents, exception-based review, and strategic judgment. Here's what the role looks like, what skills it demands, and why the transition is accelerating.

Seven AI-native ERP vendors have raised over $500 million combined in the past 18 months. Oracle, SAP, and Microsoft are racing to bolt agentic AI onto existing platforms. The architectural differences between these approaches determine what your finance team can actually do with AI, and what it can't.

The FP&A tool market has split into two tiers: platforms bolting AI onto existing planning workflows, and a new class of agentic systems that build, run, and revise financial models on their own. Here's what actually works, who makes it, and where each tool fits in a modern finance stack.

Generic AI policies don't work for finance. The regulatory stakes, data sensitivity, and audit requirements demand something specific. Here's what a real governance policy needs to cover.

Most finance teams overestimate their AI readiness by a full point on a five-point scale. Here's a structured framework for figuring out where you actually stand.

71% of finance leaders say they're concerned about measuring AI ROI. Here's a framework that ties automation gains to actual P&L impact, with benchmarks from teams that have done it.

The latest AI models don't just summarize your month-end close. They run it. A look at how Opus 4.6, GPT-5.4, and MCP-connected workflows are restructuring the finance function in 2026.

The building blocks for a compressed, AI-assisted month-end close are already shipping inside the tools finance teams pay for. Here's what a realistic 2-3 day close actually looks like.

54% of CFOs say AI agents are a 2026 priority. The competitive risk of waiting now exceeds the operational risk of deploying. Here’s how to think about agentic finance.

The U.S. Treasury and ECB both dropped AI governance frameworks in a single week. Here’s what finance leaders need to do before the next quarter close.

Microsoft is spending $37.5 billion a quarter on AI infrastructure, but only 3.3% of M365 users pay for Copilot. The structural problems run deeper than adoption lag.

Goldman Sachs, Ramp, Oracle, and HPE all shipped AI agents for finance in the same week. The agent stack just became real, and the architecture pattern is converging fast.

A guest column written by SAL, an open-source AI executive assistant running on a self-hosted VPS. Six hours to build. Here’s what works, what breaks, and what it means for finance teams.

Six weeks of dreading a messy handoff to a bookkeeper, resolved in a single afternoon with Claude Cowork and QuickBooks Online. Multiple AI tools chained across one workflow.

2025 took me around the world for AI and finance work. Every city got a run. Twenty runs across six countries and three continents, documented with Strava data and local context.