Home/Guides/Custom build vs vibe coding
Deciding

Custom forwarding system vs vibe coding: what AI-built software can and can't do

AI coding tools are real, and we use them every day. Here is where they help a forwarder, and where a vibe-coded system becomes a liability.

By Rob Petti, Founder, Lading CorporationUpdated November 18, 20263 min read
Short answer

AI coding tools make a capable team much faster. They don't replace knowing how a forwarding operation works, designing for exceptions, securing the system, or supporting it after go-live. Vibe coding is a good way to build a small internal tool or test an idea. It is a risky way to build the system your shipments, invoices and customs filings run on.

What vibe coding is

Vibe coding means describing the software you want in plain language and letting an AI model write the code. People with no programming background can now build working tools. Flexport, for example, launched a 90-day course in 2025 teaching non-engineers to build workflow automations this way (WebProNews), and operators are building API-driven tools for their own teams in minutes (Container Bytes podcast).

Forwarders should take advantage of this. The rest of this guide covers what is worth building that way.

Where it works well

  • Internal tools: a rate comparison sheet, a document checker, a report your team runs every Monday.
  • Prototypes that show what a process could look like before anyone commits budget.
  • One-off automations with a person checking the output.
  • Experiments that test whether an idea is worth building properly.

Speed, a close fit to your own workflow, and cheap experimentation are the real advantages (Magaya).

Where it breaks for a system of record

Exceptions take most of the design work

A demo handles the easy path: one shipper, one consignee, one mode, one currency. A forwarding system also has to handle consolidations, split and part shipments, amendments after filing, credit notes, currency revaluation, accruals, and month-end close. Those cases are where most of the design work goes, and they are what an AI model can't infer from a prompt.

Integrations need upkeep

Carrier, customs, and partner connections change. Someone has to monitor them, update them when formats or regulations change, and fix them when they fail at 2 a.m. during peak (Magaya).

Security

Freight systems hold customer data, payment details, and customs filings. Research cited in industry coverage found vulnerabilities in 45 percent of AI-generated code samples, including missing validation and exposed credentials (Magaya). Without a review step, those issues reach production.

Ownership after launch

Once software becomes mission-critical, someone has to understand how it works. A system built quickly by someone who has since left becomes fragile as integrations and automations pile up.

Side-by-side comparison

Vibe-coded toolProfessionally built system (using AI)
DesignPrompted feature by featureDesigned by people who have run forwarding operations
ExceptionsHandled when someone noticesMapped and tested before go-live
TestingInformalTested against your real jobs
SecurityUnreviewedEvery change reviewed by an engineer, with access controls, logging, and backups
Audit trail and recordsOften missingBuilt in, with retention planned
SupportWhoever built itAnnual retainer, trained administrators, and documentation
OwnershipUnclearEverything custom built for you is yours

How we use AI in our builds

We use AI coding tools throughout our work because they let a small, experienced team deliver faster. Around those tools we keep the parts that make a system dependable: a design based on how your operation runs, an engineer reviewing every change, tests against your real jobs, security controls, documentation, and support after go-live. AI helps us write code faster, and our engineers remain responsible for everything we release.

Combining a core system with team-built tools

The best setup lets both approaches do what they're good at. Run your operation on a solid, well-tested core with clean data and proper interfaces, then let your team build small tools on top of it within clear guardrails. Our AI training for forwarders covers how to set those guardrails.

Common questions

Can our operations team build their own tools with AI?

Yes, and they probably already are. Give them a stable core system, read access to clean data, and a usage policy covering which tools to use and what data can go where. Keep anything that files with customs or posts to the ledger inside the core system.

Will AI make a custom forwarding system cheap?

It makes building faster, which lowers cost. The work that remains, understanding your operation, designing for exceptions, migration, testing, and support, is where much of the effort sits, so expect a meaningful reduction rather than a near-free system.

Is AI-written code less secure?

Unreviewed AI-written code often contains security flaws. Reviewed and tested code doesn't have to be. That is why every change in our builds is reviewed by an engineer before release.

Talk it through with us

Tell us the problem you'd solve first, and people who have built, implemented, and supported forwarding systems will reply with how they would approach it.

Get in touch →