AI economics
What does a completed result actually cost?
A cheaper model call can still mean a more expensive workflow. Work through a cost example that includes retries and human review, then adapt the downloadable workbook.
Ajith Vallath Prabhakar
Make better decisions about enterprise AI.
I explore how people and AI can work better together. My writing gets into the choices behind that: what a useful result costs, what to entrust to an agent, and where human judgment matters. You’ll find detailed essays, worked examples and practical tools to use in your own work.
Put the work to use
Architecture
Connecting a tool is only the start.
Follow the boundary between an MCP connection and permission to act. The diagram separates transport access from business approval, with checks for credentials and untrusted tool output.
Selected recognition
The work, beyond this site.
Quoted analysis
VentureBeat
Chain of Draft and more efficient AI reasoning.
Read the coverage →Research reference
DNV
Test-time compute, cited in its review of safe and responsible LLM use.
Read the research →Media listing
MIT CISR
Enterprise Digital Twin.
View the listing →Journal citation
World Journal of Radiology
Large Concept Models.
Read the cited article (PDF) →The questions behind my frameworks.
My Control Tower Reasoning System (CTRS) connects organizational context, decision processes and agent action.
Take a customer asking for a policy exception. The system needs the current rule, the person permitted to approve the exception, and a record of what actually happened. These frameworks examine connected parts of that decision.
Explore the frameworks- Enterprise Digital Twin
- Which facts, policies and roles apply to this decision?
- Version Drift
- What changed between the recommendation and the action?
- Decision Velocity
- How long does it take to reach an accepted outcome?
- Agent orchestration
- Which system can act, under whose authority, and with what evidence?
What I’m building
Maya: a personal agentic OS.
I’m building Maya to bring the future into the way I live and work: a continuing personal system where human judgment and AI complement each other.
My knowledge, relationships and ongoing work belong in that picture. This learning library is one part of it: somewhere to return to the papers, videos and ideas I’m working through.
Explore Maya
Latest essay · 17 October 2026
Enterprise AI in 2027: seven decisions to make before you scale
An evidence-led planning outlook for AI economics, workflow ownership, context, model choice, authority, assurance and human capability, with a decision worksheet.
Read the essay →
Selected thinking
What the EU AI Act Means for US Enterprises with European Exposure
A practical review of EU exposure, current applicability dates, and the decision records US AI teams should build.
Enterprise AI has a measurement problem
Measure the complete workflow: accepted outcomes, waiting time, review effort, cost and control failures. A worked example shows where faster AI does and does not help.
Enterprise Digital Twin: give AI the context a decision needs
A purchasing example shows how current roles, policies, operational state, decision history and constraints change what an AI system should recommend or do.
Enterprise AI governance: make the controls reach the work
Connect AI policy to decision rights, current evidence, action permissions and recovery. A practical operating design and test plan for one enterprise workflow.
Start a conversation.
Bring a decision you’re working through, an idea you want to explore, or an opportunity where we could work together.
Advisory & research collaboration
AI strategy, architecture, research and writing we can explore together.
Speaking & media
Keynotes, executive briefings, interviews and conversations about the work.