Guides on Dr. Hitesh Sarda

The real cost of AI: tokens, models and who pays the bill

Mon, 01 Jan 0001 00:00:00 +0000

Every few months a lab cuts its prices and someone in the room concludes that AI is getting cheaper. Then the quarterly bill arrives and it is bigger than the last one.

Both things are true. Token prices fall, and spend still rises, because cheaper work gets used for more things: agents that loop for longer, tasks nobody would have funded a year ago, and routing layers that move the cost somewhere it is harder to see. The cost of AI is set by engineering choices (caching, model choice, how much scaffolding you send with every call) far more than by the price list.

Deploying AI agents safely: autonomy, permissions and blast radius

Mon, 01 Jan 0001 00:00:00 +0000

For most companies the question is no longer whether to use AI agents. The harder questions are how much to let them do on their own, what they are allowed to touch, and how you would know afterwards what they did.

The posts here come at that from a few directions. Some deal with reliability and why agents stall on long workflows. Some deal with permissions, sandboxes and the infrastructure that has to enforce them. A few are about the less technical side: an agent with no reputation to lose behaves differently from an employee, and the scaffolding you wrote for last year’s model may now be working against you.

Leading engineering teams that code with AI

Mon, 01 Jan 0001 00:00:00 +0000

When AI writes a large share of the code, the work of an engineering team moves. Producing code stops being the scarce part. Reviewing it, understanding it and deciding what the agent should prefer become the job.

That shift shows up in places leaders do not always look. Review queues grow while output charts look healthy. Old, accepted risks in the codebase get more expensive because AI makes them easier to find. Release dates slip because bugs are found faster than they can be fixed. And an agent makes stack and library choices at machine speed, often from defaults nobody in the company chose.

AI strategy for leadership teams: pilots, build versus buy and moats

Mon, 01 Jan 0001 00:00:00 +0000

The questions that decide whether AI pays off are rarely about which model is best. They are about whether the pilot actually worked, whether to build or buy, where to start, and what will still be an advantage in two years when everyone has the same models.

My short answers: run pilots that measure the tool and not the expert using it, start where checking the output is cheap, buy before you build, and assume the model is a commodity. The posts below make each of those arguments properly, with the evidence behind them.

AI in retail and loyalty: when your next customer is an agent

Mon, 01 Jan 0001 00:00:00 +0000

I spent years building loyalty and marketing software for retailers, and the problem has always been the same: owning the relationship with the customer when someone else sits in between. After Covid that middleman was the marketplace. Now it is starting to be an AI agent that shops on the customer’s behalf.

That changes what a retailer has to be good at. An agent does not respond to a well-designed app screen. It reads data, compares prices and policies, and picks. The posts here look at what that means for retailers, for payments and marketplaces, and for loyalty programs, plus one older post on the multi-channel loyalty problem that started this line of thinking.