Custom softwareWholesale and manufacturing
An order desk built twice, from both sides of the trade
Transcend Ltd, a UK wholesaler in perfumes and FMCG
13 implementation items assigned after the demo-kickoff review
Client projectCustom softwareAutomationMarketing and sales
An AI agent scope review that put validation before a larger build
1888Driving / Distance Learning Center, before an AI agent build.
- atomic requirements across the proposed AI agent scope
- 63
- the first Google Places allowance in the validation review estimate, not actual spend
- 4.5 to 8 times
- architecture decisions recorded in the validation review
- 15
- risks identified before a build could proceed
- 11
Proof
See the AI agent scope review for yourself
1 line from the client.
Said1
Faizan, This is a beautiful proposal. You do amazing work.
AI agent scope review at a glance
- Challenge
- 1888Driving's proposed marketplace combined an AI avatar, listing collection and automated quote calls, with scope changing across several calls.
- Solution
- We documented 63 requirements, 15 architecture decisions, 24 open questions and 11 risks after the discovery workshop.
- Result
- Close-out discussions put a validation-sprint model on the table instead of claiming that the proposed AI agent had been built.
The AI agent scope review challenge
The proposed marketplace would match consumers with local shops. Its scope combined an AI avatar, listing collection, automated quote calls, a phone menu and cloud functions. The homepage avatar was missing from one version of the scope, although the client considered it central to the idea.
What went into the AI agent scope review
Discovery and architectureAI and LLM toolsWorkflow automation
A revised January 2026 scope for the consumer app, AI calling, lead distribution, shop portal and admin area, with a written list of exclusions
The planned shop-listing collection through APFI, with Firecrawl as a fallback, and outbound quote calls through Retell AI
A Twilio phone menu, with AWS Lambda functions and SQS queues between the parts
More of what we built4
Scope calls in January, March and April 2026 as the avatar and outbound calling requirements were reconciled
An April 2026 proposal for a six-week, 360-hour AI agent integration, followed by Discovery Workshop #1 on 6 May
A review of 63 requirements across seven modules, 15 architecture decisions, 24 open questions and 11 risks
A 1 to 2 June close-out that put a validation-sprint model on the table for future work
AI agent scope review results
The scope ended with a validation-sprint model for future work, rather than a claim that the proposed AI agent had been built.
Done again today
What this would cost today
Done again today, work like this usually costs $2.5K to $5K (Two-week prototype). The free plan says whether it fits before anything is priced.
Two-week prototype
A working version in 2 weeks, on real data and clickable. The repository is yours.
Get a quote for this$2.5K to $5K, usually fixed-price
2 weeks
AutomationWholesale and manufacturing
Order emails that turn themselves into quotations
An Accel Automation presales prototype
3 quantities per stock item: on hand, available and reserved
Custom softwareSoftware and IT services
A remote-desktop relay a Toronto IT firm saw run, then funded
A prototype we built for Selenium Group, an IT-services firm in Toronto
A React portal, a .NET and PowerShell Windows agent and an ASP.NET Core relay for managing remote PCs, which the firm funded to continue and then took through a design phase.
Your project, next
Have a problem in this shape?
Say so in a few sentences. The free plan comes from the person who built this one.