The Age When AI Asks, “Does That Estimate Really Take Three Days?”

While scrolling through social media, I came across an idea that caught my attention.

In the future, it may become standard practice for AI to analyze software development estimates and ask, “Is this amount of work reasonable for this task?” or “Is this much more expensive than other vendors?”

Along with an RFP, or request for proposal, vendors would submit a detailed breakdown such as five person-days for design, ten for implementation, and six for testing. The client would feed estimates from several companies into an AI and compare them with figures from past projects and other vendors.

“Ten person-days is too much for developing this API.”

“The average among other vendors is 6.2 person-days.”

Comments like these could come back within seconds.

For clients, this sounds like a dream.

For development companies, not so much.

Estimates Were Never That Scientific to Begin With

I have worked in software development for a long time, and estimates are surprisingly human.

Even after breaking a feature down and adding up the required work, a great deal ultimately comes down to experience and intuition.

“The requirements have not been finalized yet.”

“Something is likely to go wrong with the external system integration.”

That is why a task that looks as though it will take three days may be estimated at five. The sales team may then adjust the figure further to account for risk and profit.

It may look opaque, but that ambiguity has also served as a shock absorber for projects.

After all, in software development, things are more likely not to go according to plan.

AI Will Ask, “What Are Those Extra Two Days For?”

Things get interesting when AI enters the picture.

Until now, saying, “This feature will take five person-days,” may have been enough.

But an AI with access to a large volume of historical data could say:

“The median for similar features is 3.1 person-days. Please explain the remaining 1.9 person-days.”

It would be like gaining another unpleasant boss.

And AI never gets tired. It can inspect every line of an estimate containing hundreds of items and identify that “this item alone is 32% more expensive than other vendors.”

Every part that used to pass with a vague “That sounds about right” would now require an explanation.

Then Who Takes On the Risk?

But what would development companies do if AI systematically stripped every buffer out of their estimates?

They would probably say:

“In that case, we cannot take responsibility for delivering the completed system at that price.”

Under a fixed-price contract for custom development, the development company assumes considerable risk. If it accepts a project for ¥50 million, or about $333,000, it cannot simply charge more because the work took longer than expected.

That is why risk is factored into the estimate from the beginning.

If AI labels even that risk premium “too expensive” and cuts it, there is no reason for a development company to accept a fixed-price contract.

Upfront Fixed-Price Estimates May Become Less Common

Seen this way, AI assessment of estimates may lead to more than lower profit margins for development companies.

Upfront fixed-price estimates for custom development may themselves become less common.

Instead of promising at the outset, “We will complete this system for ¥50 million,” contracts may shift toward a time-and-materials model based on assigning a certain number of engineers for a certain number of months.

If the requirements change, the schedule changes. If problems arise, the amount of work increases.

In other words, uncertainty would be built into the contract from the start.

AI may try to make estimates more scientific, only to bring us back to the question:

“Is it even possible to predict future development effort accurately at the beginning?”

Clients May Not Always Hold the Stronger Position

There is another assumption behind this idea: that clients are in a position to choose among development companies.

Recently, I heard that a company run by someone I know was struggling to find a vendor despite wanting to build a system.

With engineers in short supply, development companies can choose their projects too.

“Another company is cheaper.”

“Please reduce the estimated effort here.”

Even if a client says these things, a development company can simply respond, “Then please hire that company,” if the terms are unattractive.

The question is whether that “cheaper company” actually exists.

Even If AI Negotiates the Price Down, Who Will Build It?

With AI, it will be easy to produce impressive analyses claiming, “This phase can be reduced by 20%.”

But if you take that analysis from one development company to another and each says, “We will not accept those terms,” that is the end of the story.

A system is not completed by optimizing its estimate. In the end, someone still has to build it.

So while AI may scrutinize estimates in ever greater detail, development companies may respond, “If you want to negotiate every item down that far, let us use a time-and-materials contract,” and hand the risk back to the client.

AI will tell you, “This estimate is too high.”

It is probably right.

But AI will not:

bring you a company that will actually build the system for that price.

As long as someone is willing to negotiate, it is a price negotiation.

Once there is no one left, you are simply talking to yourself.