AI Agent Developer for Personal Automation System (Long-Term)

A bit about me: I'm a UK property developer involved across the whole cycle — buying, securing planning, building the houses (we are builders also so pricing matters), and selling through an estate agent while staying hands-on with the sales and legal side. I also run a travel content channel on socials where I visit remote and generally dangerous locations. I want a personal AI agent system, with Telegram as my main way in — I fire off a request from my phone and the agent goes away, does the legwork, and reports back — but that's only one trigger among several. My examples are the FLOOR, not the brief. I want you to show me a ceiling I can't yet see. To be clear about what I mean by "agent": I'm not after a chatbot that answers messages. I need something that takes a goal, works out the steps itself, uses tools, runs unattended for twenty minutes or two hours, copes when something doesn't go to plan, and comes back with a result. That said, I don't want everything made agentic for its own sake — some of what I've listed is simple, repetitive and should be built as plain deterministic automation. I'd value your view on which is which. The kind of work I have in mind: • Recurring emails handled so I never see them: the attachment extracted and filed, the email archived, downstream jobs picking it up from there. • Filing, which is my single biggest time drain. Documents arrive constantly — drawings, reports, quotes, council letters. I want them read, correctly named (a prefix added, never overwriting the sender's own filename) and filed into the right folder across six live developments, without me touching them. • Planning condition management. A decision notice arrives listing conditions; I want the tracker and folder structure built from it, each condition chased with the council on a ~10-week cycle, and the discharge letters filed and status updated when they come back. • Form-filling into a government planning portal — a repetitive ~10-step submission (document uploads, references, site coordinates, drawing a site boundary), triggered by a status change in a tracker spreadsheet, with the result written back to the sheet. • A quarterly supplier pricing round: emails out to our merchants, replies gathered into a comparison spreadsheet. Suppliers name the same product differently, but we hold a year of historic price lists to match against. • Competitor market watch: every house for sale around our sites recorded and compared against our own pricing month to month. • Scheduled jobs that edit existing live spreadsheets in place, e.g. a monthly cashflow forecast built from figures our admin team enters. • Errand-class tasks dispatched from my phone: "get me quotes for two front tyres and lead times", "find me a table in Belfast Friday around 7-8". Agent researches or phones round, comes back with options, acts once I approve. • Document retrieval on demand — I ask for a file, it finds it across OneDrive/Drive/local storage and emails it to me. Same for pulling the latest report off a public planning portal, where the file is rarely named for what it actually is. • Publishing: drafting and posting blog content to WordPress. Voice calling (an agent that phones a business, asks intelligent follow-ups and returns a transcript) is a near-term ambition rather than day one. On the model: I use Claude and rate it, so that's my starting assumption — but I'm not committed, and I'd rather you tell me what you'd actually use and why. Different tasks may warrant different models. What I do want is an architecture I'm not locked into, so a better model can be swapped in later. IMPORTANT — this is an ongoing responsibility, not just a build: I want someone operating at the edge of what's being released in AI, whose job includes proactively telling me when a new model, tool or capability is worth adding to my system or opening up a workflow I couldn't run before. Not novelties. If you're not actively tracking this space week to week, this isn't the role for you. I'm hoping to form a lasting relationship rather than run a one-off project. Please answer in your proposal: 1) Give me three specific things you believe an agent could do for ME that I probably haven't thought of, tailored to my world above. No generic one-liners ("an AI fitness coach", "create a development appraisal for a new site") — I want a worked example per idea: what it does, how it plugs into my life, and the concrete payoff. Depth will win me over :) 2) Tell me about something you've built that ran unattended — and what it did when a step failed halfway through. 3) What do you think the hardest part of this will be, and what would you do differently from what I've described? Which items on my list would you NOT build as agents? 4) Would you build one general agent or several specialised ones, and why? (I mean the technical structure of the AI system — I'm not in building architecture.) Which model or models would you run behind it, and how would you keep that swappable? 5) What's a recent AI capability you've adopted and why — and one you deliberately chose NOT to adopt, and why not? I'll shortlist a handful of people, send you a detailed brief on what I'm trying to achieve, and pay for a consultation call to go through your ideas in depth. So please make sure your proposal shows you've genuinely read the above.