AI agents are making their way into the design of route planning tools. According to a survey conducted from 1 July to 8 September 2026 by Kardinal, a route optimization specialist since 2015, 35% of the developers, CTOs and tech leads surveyed already use an AI agent daily. Their main challenge, however, has not changed: for 56% of them, planned routes fail to match real operating conditions, whether the tool is built in-house or not.
AI agents are entering tool design
Teams that build planning tools are changing the way they work. One respondent in three (35%) uses an AI agent daily for development, and 23% occasionally. Only 17% use none. Claude Code is by far the most cited tool, ahead of GitHub Copilot, and most users combine several tools.
These teams also expect more transparency from their vendors: nearly two thirds (65%) are strongly bothered by a tool without public pricing, and 26% refuse to even evaluate it. Their top concern remains a bill that could balloon (41%).
The gap with reality, the top challenge
The gap between planned routes and real-world conditions tops the list of difficulties (56%). It ranks ahead of performance and compute time (47%) and constantly shifting business needs (43%). Nearly four respondents in ten (39%) even struggle to know whether a planned route is good. Tool cost (37%) and integration into the IT system (30%) come next.
One respondent sums up this blind spot: “It is hard to know whether an optimized route is genuinely better than one built empirically by an experienced driver.”
Doing it yourself does not solve the problem
In-house development remains the most common answer: 56% of teams have built their own planning tool. Spreadsheets have not gone away: nearly a third (31%) still rely on them, at least in part, as many as use off-the-shelf software or an API (31%). These approaches are often combined: nearly half of the teams that develop in-house also use another tool.
But doing it yourself does not remove the No. 1 difficulty. The gap with field conditions affects teams that develop in-house (51%) as much as the others (50%).
Maintenance weighs on teams
Among respondents who operate their own solution, 35% report regular maintenance and 18% heavy maintenance with dedicated resources. Only 12% say it requires almost no effort. Without real-world constraints, the optimal route remains theoretical
Teams aren’t short of algorithms: powerful solvers are available as open source. What makes the difference is the ability to represent what really happens in the field: delivery windows, loading docks, refrigerated vehicles, customer habits. As long as those constraints aren’t in the model, the optimal route remains theoretical
says Cédric Hervet, co-founder of Kardinal.
