I write about making AI useful in the work people actually need to do. That includes the architecture behind a dependable system, the authority it should have, and the question of whether the result is worth its cost.
You do not need to read the site in order. Choose the route closest to what brought you here. Each starts with a practical question and leads to something you can use in a discussion or test in your own setting.
I’m deciding where to invest in AI
Start with the outcome you want to improve and the evidence you would need to justify a change. A faster answer is useful only if the work that follows also improves.
AI token economics works through the cost of an accepted result, including failed attempts, review, and rework. Its editable workbook lets you compare two configurations using your own assumptions.
Buying an AI agent sets out three questions to test before committing to a service: the outcome, the evidence under your operating conditions, and the work of running it after launch. Use the evaluation workbook to record what the vendor demonstrates and what remains unresolved.
The 2027 outlook sets out seven planning decisions, from workflow ownership to model replacement and human judgment. Use the accompanying worksheet to name an owner, a next test, and an acceptance condition for one workflow.
Decision Velocity follows a refund request through preparation, waiting, approval, and confirmation. It is a way to examine where the complete process is losing time, rather than measuring the model in isolation.
I’m building or governing an AI system
Choose the problem that is hardest to make explicit: the information the system needs, the action it may take, or the evidence that would let another person check its work.
The frameworks bring those questions together with worked examples, visual maps, and downloadable worksheets. CTRS examines context and authority; Version Drift considers a policy that changes while a decision is pending; Agent Orchestration follows a purchase-order request across an approval boundary.
Personal Cognitive Digital Twins looks at what happens when remembered preferences and current intent differ. The examples are useful for testing whether a personalized assistant respects a correction rather than merely recalling more history.
Progressive Capability Discovery examines how an agent finds an appropriate tool, recovers from a poor selection, and distinguishes an available operation from an authorized action.
Get to know me and my work
Maya is the personal agentic OS I’m building to bring my reading, conversations, and ideas into a more connected way of thinking and working with AI.
You can read about the curiosity that started it all in About Ajith, or find speaking topics and selected references to my work on Speaking & Media.
If something here connects with a question you’re exploring, let’s talk. Tell me what you’re working on and what you’d like to explore together.
Bring a useful tool into your next discussion
The enterprise AI working brief is a four-page starting point for a team conversation about context, authority, action, and evidence. It includes a visual map and space for working notes.
The AI cost workbook compares two AI workflows with a manual baseline across four sheets. It includes model routing, external APIs, human effort, operating costs, and volume analysis. Replace the illustrative inputs with your own rates and task outcomes.
The 2027 decision workbook has five fillable pages connecting the seven decisions to process redesign, complete operating costs, action authority and investment gates. The personal OS workbook has four fillable pages exploring intent, memory, permissions and the work that returns to a person.
AI supplier evaluation workbook
All five resources are available without signing up. The linked articles explain the assumptions and the reasoning behind them, so the downloads remain connected to the work they support.