My view on the market, finance ops and where AI agents fit. Each essay makes one argument and shows what it rests on.
Can an AI agent reconcile accounts and explain variances? It can match, propose fixes and draft notes with source links. A person still signs the reconciliation.
Read What a Reconciliation Agent Can Explain, and What a Person SignsCan an AI agent offer a payment plan or take a promise to pay? Yes, from your policy table. Here is when it stops and hands the account, or the call, to a person.
Read Can an AI Collections Agent Offer a Payment Plan, and When Does a Person Take Over?Margin-based pricing sets the price of an AI product from what each customer costs to serve, not from what the market will bear or what the model bill says in total.
Read Margin-Based Pricing for AI ProductsYes, under conditions. An agent can pay a supplier invoice through a portal when a person approved the invoice first, the agent carries a credential that works for that payment only, and the portal knows which agent it is talking to.
Read Can an AI Agent Pay a Supplier Invoice Through a Portal?Paddle's AI Toolkit puts billing setup inside Claude Code, Codex and Cursor. That moves price changes into pull requests, and finance needs a seat on the review.
Read Billing Setup Just Moved Into the Coding ToolFour AI-native revenue systems sold to payments, CRM and accounting giants in ten months. Why the buyers are paying, and what it means for founders and finance teams.
Read AI-Native Finance Systems Are Getting Acquired EarlyCan an AI agent apply cash and post to your ERP? Yes, inside gates. Here is where the human gate sits for tolerance, short pays, deductions and low-confidence matches.
Read AI Cash Application: What an Agent Can Post and What It Must EscalateFor four weeks, a crew of Grok bots ran a small online store. Here is what they built, what the numbers showed, and five things I did not expect.
Read I Ran a Four-Week Experiment With a Crew of Grok Bots. My LinkedIn Reach Tripled.My view is that stop conditions are the product, not a safety review after launch. A stop condition is the rule that tells an agent to halt, escalate, or hand off to a person.
Read Stop Conditions Are the ProductAn AR agent reads invoice state, decides the next action under finance policy, acts in your systems, and follows through until cash posts. It is not a chatbot or brittle RPA.
Read What Is an AI Agent for Accounts Receivable?The Machine Payments Protocol lets agents pay. It does not let them buy. An invoice is credit, not a request for payment, and agentic commerce will arrive through accounts receivable.
Read Agents Can Pay. They Can't Buy. An MPP Update?Pull what has a system of record. Persist what does not. A folder is not a memory, and most of the knowledge that matters has no system of record at all.
Read Your Agent Needs Two MemoriesMCP answers how an agent reaches data. The brain answers what the organization knows. Pull facts of record. Persist judgment that has no system of record.
Read Facts Flow, Meaning AccumulatesThe companies stuck at Assist-level AI aren't stuck because the technology plateaued. They're stuck because nobody is doing People Operations for the agents.
Read Your AI Agents Need an HR DepartmentThe software meltdown isn't AI killing SaaS. It's Christensen's Law of Conservation of Attractive Profits playing out in real time. Value is migrating from the interface layer to the context layer.
Read The Context Layer Is Eating SaaSKarpathy described a personal LLM wiki. We have been running the multiplayer version since the start of the year.
Read The Shared BrainWhat does it actually mean for a document to 'decide' something? The four components that give an invoice or contract genuine agency.
Read The Anatomy of a Document AgentFinance leaders are deploying AI agents like they're adopting teenagers. Curfews, check-ins, no real decisions. Then they wonder why the agents don't deliver.
Read Why Your AI Agents Are Failing: The Control TrapEveryone agrees on where AI agents are going. The destination is not the differentiator. The route is. Constraint profiles, a task vocabulary, and democratized creation.
Read The Execution Layer: Why AI Predictions Are Just EntertainmentMost finance teams deployed AI and got incremental results. The problem isn't adoption. It's that no one specified which kind of AI to adopt. AI comes in four kinds, not one.
Read The 75% of AI Value You're Leaving on the TableWe've been thinking about AI in finance backwards. The end state isn't smarter agents processing dumb documents. It's documents that process themselves.
Read The Invoice as Agent: Why Every Document Will Have Its Own AISixty percent of enterprises require human oversight before AI agents access sensitive data. It feels responsible. It's a placebo with a 24-month shelf life.
Read Human-in-the-Loop Is a Temporary Placebo
