The excitement around AI and the return from using it in everyday work are different things.
I care more about a concrete result: can a business card become a customer record that we can continue using? Does an outreach email have factual support? When the next customer arrives, have we retained the reasoning from the last one?
Return to the work of export sales
Product descriptions, images and copy are accessible starting points. Customer development quickly encounters judgment: is the person behind this card genuine? Is an inquiry a real requirement or a price comparison? Would this quotation lose money?
A poor judgment can cost an order or produce a wrong price. Even when AI drafts quickly, the salesperson must check whether each claim has evidence and whether each advantage rests on concrete facts.
The time spent polishing a draft reminds me that generating something is not the same as completing the job.
The problem is not always the tool
I made this mistake myself.
AI wrote some promising outreach emails, so I immediately added research, quotations, follow-up and pro forma invoices. After a few days, the whole chain broke down.
A narrow task works, and then we ask the system to take on adjacent work. Every extension adds an integration, a permission and more places for errors, without necessarily adding the foundations to support them.
Another familiar problem is working inside a conversation and leaving everything there. The next customer starts from zero. Three months later, we cannot explain why a prospect looked promising or why we offered that price.
Both problems point to the same question: what exactly should this piece of work deliver?
Without an agreed output, scope has no boundary. Without an inspection standard, “looks good” becomes the only measure of completion.
Say where you want to go first
One of my four principles for using AI is to plan for a concrete outcome.
Have an intention and a plan. Know where you are going and what result you want.
Say where you want to go before debating how to get there.
When the objective is vague, the assistant has to guess. More speed and capability can still produce wasted effort.
The objective must also be inspectable. A customer record needs sources for its facts and explicit unknowns. An email must remain a review draft, with recipients, facts, attachments and the signature checked.
How an earlier product revision broke the goal into steps
When I first published this article on October 20, 2026, I had narrowed the website from several template entries to one export-sales workflow. Version V0.4.2 used a downloadable materials pack, a local workspace and six steps.
The steps started with a business card and meeting notes: confirm the inputs; organize facts and a customer record; plan communications; prepare an English first draft; process a simulated reply; and let a person decide about quoting, sending and ownership of the next action.
The important part was not the number of steps. Each step said what we needed to see before continuing.
- Facts needed sources, and unknowns could not be invented.
- Emails stayed unsent, pending checks of facts, attachments and signatures.
- “Possibly 24 units” was not an order or a confirmed quantity.
- Status had to match the person’s decision, and the work had to leave readable records.
The agreed outputs then included company and product facts, a customer archive, a communication plan, an outreach review draft, an inquiry analysis and reply draft, and a requirement clarification and status record. These were inspectable results, rather than evidence merely that someone had used AI.
In a new conversation, the assistant could read those saved records and explain where the work stopped, what the person decided and what information was still missing. That is what made continuation possible.
The product has changed; human decisions remain
The description above concerns the earlier materials pack, not current installation instructions. Today the entry is the Broadlatus desktop app with the Trade Standard Skill. See the current download page for the actual release and scope.
The skill supports customer records and research, outreach strategy, email preparation and independent review. The workbench brings customer decisions together so a person can inspect evidence, record a decision, wait for assistant pickup and review the result. See the workbench user guide for setup, operating requirements and limits.
The earlier basic version did not research customers online or send email. Even with research collaboration now available, retrieved information is not automatically a verified fact. Sources, gaps and human checks remain necessary. A fluent paragraph cannot create an approved quotation, a delivery commitment or permission to send.
First deliver something that can be checked
Completing a workflow does not guarantee income. Its value must be judged against your own tasks, effort and results.
Choose one piece of export-sales work you repeatedly start from scratch. Decide what it should leave behind, when it must stop and ask a person, and who owns the next action.
Define the deliverable first. The rest follows from that.
Share specific obstacles through our feedback and co-creation entry. A concrete problem helps improve the product more than “looks good.”