Disclosure first: I'm a co-founder of Kardinal, a route optimization API. We ran this survey to understand how technical teams actually deal with routing problems. Some results confirmed what we expected, others didn't, so I'm sharing the data with its limits.
Methodology (and its biases)
- 145 respondents, collected July 1 to September 8, 2026. The results below are restricted to the 99 technical profiles: developers/engineers (34%), CTOs/heads of engineering (27%), tech leads/architects (18%), other technical roles (20%).
- 39% work at logistics/transport operators, 36% at software vendors/SaaS.
- 76% work on route optimization regularly or occasionally.
- Each question has its own base (conditional questions, skipped answers). I give it every time.
- The survey was distributed by Kardinal, so our customers are probably overrepresented. Read the numbers as signals, not as industry truth.
1. The No. 1 difficulty isn't compute time
"What was the hardest part?" (base 89, multiple choice)
- Results that don't match real-world conditions: 56%
- Performance / compute time: 47%
- Shifting or hard-to-translate business needs: 43%
- Hard to tell if a result is good: 39%
- Cost / pricing: 37%
- Technical integration: 30%
The cross-analysis is the interesting part: teams that build in-house face the real-world gap just as much as the others (55% vs. 58%). Building it yourself doesn't remove the No. 1 problem. And in-house teams report more performance issues (53% vs. 37%) and more integration issues (35% vs. 24%).
One free-text answer summed it up better than I could: it's hard to know whether an optimized route is actually better than one built empirically by an experienced driver, and metrics of gap-to-reality would be more useful than purely algorithmic ones.
2. Build is the default, and maintenance doesn't stop
56% build in-house (base 99). A third of respondents combine at least two approaches (in-house + third-party API, manual planning, or open source). The most cited open-source solvers: OR-Tools, VROOM, Timefold, pgRouting.
Among those who operate their own stack (base 60): 35% report regular maintenance and 18% heavy maintenance with dedicated resources. That's half of them.
Self-hosted / open source is still the preferred model for 45% (base 80), ahead of a managed API (31%). 24% say "it depends".
3. AI agents are already part of the workflow
Base 92: 35% use AI agents in their daily workflow, 23% occasionally, 20% have experimented with nothing in production, and 17% don't use them at all.
Among agent users (base 83), Claude Code is cited by 81%, GitHub Copilot by 45%. Multi-tooling is the norm: 65% name at least two tools.
This changes what "good docs" means. Clear documentation was the No. 1 adoption criterion (top 3 for 59%), followed by code examples (58%) and guidance on modeling the problem. An agent will read your docs literally and won't ask when something is ambiguous.
4. Hidden pricing costs you evaluations
"An API with no public pricing, where you must contact sales: how much does that bother you?" (base 92)
- Bothers me a lot, but would still evaluate: 39%
- Dealbreaker, wouldn't even evaluate: 26%
So nearly two-thirds are strongly put off. Software vendors are much more sensitive than logistics operators (76% vs. 53%).
Top pricing frustrations: a variable bill that could spike (41%), service cut off mid-month (37%), having to watch usage to switch plans (35%).
Full results, all 15 questions and the PDF report: kardinal.ai/survey-api-2026
What I'm curious about
If you've built or integrated routing optimization (OR-Tools, VROOM, Timefold, a commercial API…): what broke first in production? And how do you measure whether an "optimal" route is actually better than what your drivers would do?
