November 16, 2026
3
min

Before the code, the conversation

Tom Cools
Keep up with Timefold
Product updates, optimization insights, and stories from the builders behind PlanningAI.
Share

The easy part, apparently, is the code

Writing code has become the easy part. As a software engineer, I'm still sometimes surprised I can write those words without feeling I'm exaggerating. I've watched AI build a working optimization model, tests and UI included, in a couple of hours (with the right building blocks).

So it's no wonder that when a team sets out to automate their planning, most of them reach for the keyboard first. Open an editor (or agentic terminal), write a model, hook up some data, done. With AI writing a big chunk of that code, the urge to just start building has only gotten stronger.

But here's the thing: AI can't tell you what you should have asked it to build in the first place. Figuring out what to build is still the hard part, and it's the part that decides whether the result is any use at all.

Yet most of our software industry seems focused on the other side of the problem: output, tokens, and "writing better code". We see software factories, harnesses, and who knows what the next buzzword will be. They all promise to build the perfect code faster, but very few help you figure out what that code should actually do.

The real rules live in people's heads

In our December newsletter, we compared integration work to cooking, where some of the ingredients have to be foraged from the brain of that one planner who "just knows". Requirements are no different. The rules that actually make a plan usable are rarely written down.

Reading through the problems people described to us over the last few weeks since we’ve launched our custom model builder capability, we came across gems like these:

  • A loaded gas ship sails 20% slower;
  • A nurse in training still covers half of a pediatric workload;
  • Customers carry each other's paragliders to the repair shop, and get ten euros off for it;

Off-the-shelf tools rarely capture details like these. Yet leave just one of them out and you end up with a beautifully optimized plan that nobody can actually use. Worse, some of these requirements force a significant redesign of the model, so the earlier you surface them, the better.

A conversation partner that asks the annoying questions

For our own AI solutions, we wanted to flip the order: get the requirements right first, and help people talk through their problems before anything gets built.

You describe your scheduling, routing, or resource allocation problem in your own words, and our Assistant keeps asking questions until it has a clear picture of five things:

  1. What the plan decides.
  2. The hard rules, which can never be broken.
  3. The soft goals, which you want to improve but can bend.
  4. The scale, for a normal day and for your busiest one.
  5. The data that goes in and the plan that comes out.

Our assistant is a bit like that colleague who keeps asking "but what about X?" or “why this?” until you're secretly glad they did. More like a coaching trying to get the best out of you. It doesn't sound as glamorous as "writing a bunch of code"... but we feel it hits the right notes.

This definitely doesn’t mean we need all requirements perfect and finalized up front … that is never the case anyway. We do however firmly believe that talking through requirements first avoids a lot of iteration hassle.

Talk first, type later

Writing code keeps getting cheaper, but knowing what to write hasn't gotten any easier. Sure, you can use an iterative approach to improve a solution. But getting that first solution close to what you want does require some back and forth with someone who can act as your sparring partner.

So before you open that editor or send your AI agent off to build, have the conversation. Get your requirements at least a bit clearer before you start coding, because coding really is the easy part now.

‍

Related blogs
October 12, 2026
0
min

Technician scheduling software in 2027: Essential capabilities for modern field service

FSR
October 7, 2026
0
min

Field service scheduling: how to balance jobs, skills, SLAs, and travel

FSR
September 23, 2026
0
min

How System One Models like Jev complement planning optimization

Optimization
Scheduling intelligence
August 10, 2026
0
min

MTTR, CSAT, SAIDI, and SAIFI? How scheduling and routing optimization drives customer satisfaction in field service

FSR
Optimization
Scheduling intelligence
July 30, 2026
0
min

SLA compliance: What missing deadlines actually costs you in field service

FSR
June 25, 2026
0
min

Same coffee, same shoes, same plan: the beauty of no surprises

Optimization
Explainable AI

When scheduling works, everything works.

Less waste. More control. Teams that trust the plan.