Speaker: Tony Galati
For more see the table of contents
Intro
- A Product Owner (Alex) at NAIC (National Association of Insurance Commissioners) gave the background of their project.
- Origin: Gave a prompt for the backend/infrastructure and have Cursor generate the draw.io. Included Okta, Docker, etc. Showed the prompt. It’s about 20 lines and pretty detailed
- Used prompts to make a front end prototype. PO iterated on it.
- Then Tony spent about 2 days connecting them.
- After that, they did a two day hackathon because knew could get something up quickly
- At hackathon 1, learned need to have business problem, implementation plan and business support. Did two business days (9-4 each day)
- Doing second hackathon next week.
- Did daily standup at hackathon
- Made sure had everything needed like Okta in advance of hackathon
Business benefits
- Can generate use cases
- If want specific format say it
- AI rewrote his one sentence prompt to include depth needed
- Got Next.jS front end prototype.
- Business can iterate with prototype independently
- Showed user stories generated
- Even if don’t download code, will speed up time for business analyst
- Figma.AI doesn’t let you change with prompts after initial generation. V0 lets iterate.
Back end
- Switched speaker to Tony – Enterprise Architect
- Absolutely not production ready, but could show working
- Can specify to AI tool what coding standards to follow, needs to work on all devices, etc
- Used Amazon Q first. Then started using Cursor
- V0 is good for live changes in front of customer.
Before Hackathon
- Engaged security and legal. AI acts on behalf of user so has user permissions.
- Went through what the models were trained on: https://trust.cursor.com
- Elaborated on AI policy. He did a show of hands and about 2/3 of the audience has an AI policy at work
- Defined intellectual property allowed to be used in Hackathon. Could have called it a POC.
- No PII data
- Once commit code to Git, normal software development lifecycle.
- Security engineer paired with developers to understand what Cursor can do
- Setup Cursor IDE Project Rules – written in plain English. Had AI write it and human proofread. Can be context specific so can say some rules only apply when you say “commit” or other scenarios
- Setup memory bank – includes extra info/tasks
- Setup pipelines and quality gates
- Wrote team prep instructions. Keep it short
- Wrote down the tech stack. Team implementing used Angular so agreed to let Cursor translate.
Future
- Roll out Cursor
- Second hackathon
- Smaller consumable guide/instructions
- Mandatory walkthru sessions including BFF (backend for front end) pattern
- Three days instead of two. This time using a language they aren’t familiar with instead of Angular.
- AI first behavior – AI does a lot – ex: write tasks for AI consumption but human readable, our job will change towards steering AI. Have AI do one step at a time
- No training because changes so fast. Instead pick champions and three day immersion
- Buy things a year at a time since change so fast
Further future
- AI code reviews.
- Have multiple agents fix a defect and primary recommend what best
- Future problem – new people won’t know codebase. Already have that problem and have to figure it out but won’t be worse. Will be catastrophic failures.
Key takeaways
- Encourage staff to use AI – even at home. You need to fix the toilet, use AI
- Start POCs
- Visuals sell
- AI is coming. Need to figure out how our jobs will change
My take
The speaker on stage was wearing a suit which made me nervous this wasn’t going to be technical. But he quickly said he was going to give an overview and turn it to Tony. Tony is wearing sneakers and jeans which is in keeping with what the hands on folks wear to conferences like this. The speaker in the suit asked how many people from the business side were in the audience. He made a joke that was that he expected when their were crickets. The information in both halves of the presentation was great. Excellent end oot the day. I’m glad the conference organizers gave them a big room! I also like that it was a realistic description and not “see AI is magic and does everything by itself”