LMCane said:
I still don't see how in the near term even a Claude or CHATGPT 6.0 will be able to:
find commercial invoices from the parent company in Israel. understand the transaction. understand whether to use a license exception or license exemption or DSP-5 or DSP-73.
go into the US department of state automated system (which requires identrust certificate) and create a license, then review the license, then submit the license (when all submitters must be a qualified Empowered Official under ITAR 120.67)
then arrange with the freight forwarder the shipments, then discuss with headquarters changes in the transmittal letter.
there are so many moving parts to every single day of imports, usage of DPAS authority for manufactured parts, decisions about where and how to ship-
how can AI do that now?
Long post, but want to be thorough. The reason I say this is a pretty good use case is because you have a decent process built already.
Something that has helped my thinking is not really thinking of AI in terms of "Claude or ChatGPT 6.0" but a team of specialists, each of which can be trained and equipped based on what you want it to do. This is a sketch of how I think this could be done today (obviously I don't know your business so some of this is probably wrong but I think it will give you the idea).
The first specialist we will call israel_bot.
- It's only job is to look at a group of commercial invoices, pick out the ones from Israel, and then hand the Israeli invoices to another specialist.
- In order to do this, we will equip israel_bot with a playbook. This playbook tells it how it would distinguish an invoice from Israel from other invoices. So imagine if it was a person and you were creating a binder that had detailed instructions for how you do this.
- That binder you made for a person might include info like "if you have trouble with deciding whether it's from Israel or Jordan, reference this documentation in the appendix (or guide from the internet or whatever)." You can provide israel_bot with these types of instructions that it would only reference if it needs them just like a person.
- The last part of Israel_bot's job is to hand the invoice to the next specialist, license_bot
license_bot
- This guy's only job is to determine whether this particular invoice requires an exception, exemption, DSP-5, or DSP-73. And then pass on the document to the next specialist once it's reached its conclusion.
- Just like with israel_bot, we create a playbook that instructs license_bot how to decide.
- And just like iIsrael_bot, this playbook can include additional context it might need to access in some situations but not others.
- The last step is for license_bot to indicate what category it belongs to and hand it to the next specialist.
prep_bot
- It sounds like this submission to department of state is something we wouldn't trust to an AI to do because it's either unsafe or illegal or both. So instead of having an AI that would do that work, we will just have a bot that will set the table for the human to do this work.
- prep_bot's job is to make it as easy as possible for the human to do the work in the DoS system. I don't know what that entails but I'll guess.
- prep_bot generates a document that has all of the information that is going to be needed to do the work in the DoS system.
- Based on the work that license_bot did, prep_bot knows what the license decision is and I'm going to guess that means it knows what forms are going to need to be filled in the DoS system, the data that needs to be put into those fields, phone number of the DoS help desk, links to supporting documetnation that might be needed, etc.
- Similar to the other bots, you just write prep_bot's playbook to get it to serve up all of the information that the human might need.
prep_bot_2
- Whatever the outputs are from that work in the DoS system, the human can give those to prep_bot_2.
- Sounds like the conversations with the freight forwarders and HQ also need to be handled by a human. So prep_bot_2 is going to do something similar as prep_bot.
- prep_bot_2 looks at the original invoice, the DoS paperwork, and any other helpful documentation that would equip it to create a document that would have all the info they would need to have productive conversations.
Nothing I described above requires "hard plumbing" of one data system to another. The capabilities available today with off-the-shelf tools like Claude Code can fetch data from any system that is connected to your computer.
Whether you should trust it to do that is a different question. Security on this stuff is still very iffy.FUTURE IMPROVEMENTS:
- You're always going to be limited by what the tools can do and what you TRUST them to do. Both of these things are going to be moving targets.
- Maybe right now you don't trust prep_bot_2 to do anything other than get a human ready to make phone calls. But someday in the future maybe you'd trust it to draft up RFQ emails the person could send out quickly. Or maybe at some point you trust it to connect to your vendor management system, analyze recent activities from the freight forwarders, and send out the RFQs automatically.
- Someday way in the future maybe the systems are trustworthy enough to actually execute orders automatically.
- Even stuff like that "automatically execute orders" are capabilities that exist today through traditional automation. This is happening in the background every time you order something on Amazon. The difference with these systems is that you don't have to necessarily directly connect one data system to the other. Just like a purchasing manager is filling this gap for my company today, an AI specialist with the right playbook could potentially do it very soon.
