# What two people +AI can and can't deliver > A Sydney AI consultancy of two on seven recent projects, from an enterprise eCRM to a legacy .NET upgrade, and where AI did and didn't pull its weight. Source: https://www.hiaitus.ai/posts/two-person-ai-team-what-we-deliver _Published 2026-10-07_ ## Key points - Hiaitus is two people. Seven recent projects show where AI pulled its weight and where it didn't. - AI did the volume on every job (first drafts, code, testing). - The judgement, strategic thinking, security and engineering were absolutely ours. - Capacity management is likely the next key challenge. Two people can't run ten big projects at once. - We pick our work and finish it. Reputation is #1. Every second post in my feed says AI lets a small team punch above its weight, and whilst that may be true, it's also something very easy to 'say'. I'd rather show you what it looks like in the 'do' phase, warts and all. Hiaitus is Tim and me (Stu), a two-person AI consultancy and venture studio in Sydney. We've got decades each in digital, marketing and software, and we work with local and international brands and agencies. We use AI on every job, and we've used it long enough to know where it saves us days and where it would get us into a quagmire. Here are our last 7 projects, and where Human Intelligence really made the difference. ## Three automotive projects, working alongside their agency Three of the seven were for a global car brand, through its agency. The biggest was the eCRM program, covering customer journeys and the emails behind them. We used Claude specifically to chew through detailed user journey maps, segment logic and the first drafts of a lot of eCRM automation emails. Claude wasn't able to make the critical calls on where to focus the time and effort for the biggest impact, being the moments between initial enquiry and turning up for a test drive. That was all our strategy and judgement, with decades of experience leaning into the real moments that matter to get the biggest bang for the client's buck. Then a vehicle launch page, built inside the brand's existing enterprise CMS, the kind with structured content types, granular editorial roles and multi-stage publishing workflows. Tim built it with Claude and the build itself was delivered ahead of schedule (thanks to remote weekend working and a few beers). The third was a price saving calculator. Brief to working page went fast. The harder question was what the page should tell consumers, because a public savings figure needs a person and the client's legal team to sign it off. ## A legacy .NET upgrade An Australian business had an ageing .NET system. We assessed it, then took it from legacy to the latest version. AI led the code audit, reading old code and explaining it back, mapping dependencies and drafting the upgrade path and tests. Tim, an expert in .NET who hadn't used it in a while, then made the key calls on what to change, what to leave alone and when to cut over. That's senior engineering judgement. Getting his hands dirty, he took the Claude Code output and, like writing an article where AI does the first draft, used his decades of experience to orchestrate the final engineering, package it up and deploy it to the client environment on time. ### Both examples enabled us to deliver vastly different projects typically needing a team of 6+ to do, with just 2+AI (in this case, mostly the AI was Claude, Claude Design and Claude Code on Opus 5.5 Max). ## Our own two products Checked It ([www.checkedit.ai](https://www.checkedit.ai)) is our AI pre-screening platform for advertising in regulated Australian sectors, rebuilt by Tim and me on a production cloud stack. I again used Claude Design and Claude Code to create the entire front end (UX flow, copy, design, SEO etc). A team of 5+ would typically be needed, but 1+AI executed the lot for the cost of the LLM monthly subscription. We also used AI to research dense federal and state legislation and turn it into structured rules, and it helped us build and test the compliance checks. Checked It reports are indicative screening to act on before submission. They're not a clearance, and a human reviews and signs off every time. Siftly helps businesses respond to RFPs and EOIs, and the MVP is live. AI got us from idea to a working MVP quickly. Who buys it, why and what they'll pay is founder work, which we're exploring with investors today. ## The CRM we built instead of buying one We needed a CRM, so we built one. It took just 5 days, it's been in daily use since July 2026, and it runs our contacts, pipeline, activity, forecasts, scheduling, board reports and plenty more. The paid route was a leading CRM's professional tier at around $800 a month, with a 3 seat minimum. For two of us that means paying for a seat nobody uses, and about $9,600 in year one. Ours cost about $300 in additional Claude Fable 5.1 token credits. We saved about $9,300, which is a lot for a small business! ## What the seven have in common On every one, AI did the volume (drafts, code, research, documentation, testing). The Human Intelligence (arguably I have a lot less of that than most, and a lot less than Tim for certain) created the original idea based on real-world problems and pain points (only experience can really define those succinctly), and ran the program orchestration, the final judgement and the sign-off. That's the part clients pay senior people for. I mentioned in a LinkedIn article that perhaps AI has flipped the over-50s ageism of 'too old to hire' to 'too experienced not to hire for anything that has AI involved' (read it here: [LinkedIn article](https://lnkd.in/p/gs4Dc-UM)). The one finite limit is our capacity. Two people can't run ten big projects at once, whatever the tooling, so we're picky about what we take on, and we finish it. (I also wrote about why the whole agency model is heading this way in [The agency pyramid is being flattened](https://www.hiaitus.ai/posts/ai-agency-pyramid-small-teams-media-budget).) Reputation is more important to us than revenue. We know the latter will follow if we're laser focused on the former. ## How we keep 'quick' from turning into 'slop' Speed is the easy bit with AI, and we've spent significant time putting clear gates, guardrails, governance, security and quality check points into every plan, every design, every piece of copy and every build. They all run in small batches, with testing, validation, confirmation, correction and human review in between. Nothing reaches a client, or gets published on this website or posted anywhere, without Tim or me reviewing it. It's a key reason two of us can move this quickly without leaving a mess for the client's team (or us) to clean up. ## When we're a good fit (and when we're not) We suit projects that went quiet, or got overestimated in time or cost because they were a bit too small or medium for a large agency to do without a full sprint team, or that need very specialist skills, like .NET or certain CMSs and CRMs. We're also a good fit for Q4 overflow, work delivered under your brand with a clean handover, internal tools, and prototypes that need to become real production software. ## How this article was made We say AI does the first draft and people make it good, so here's the proof for this article. Claude wrote the first draft, then I rewrote it. **AI vs HI score** (word-level comparison of the body copy, done by script): - Only **33%** of Claude's first draft survived to the final - Compared with the first AI draft (v1): **72%** of the final words are new or changed by Stu - Compared with the last AI version (v5): **46%** of the final words are new or changed by Stu - The article grew from **840 to 1,190 words**, and nearly all of the extra 350 are Stu's. _FAQ_ ## Questions we get a lot ### What can a two-person AI-assisted team deliver? More than most people expect, within limits. Recently the two of us delivered an enterprise eCRM, a launch page on an enterprise automotive site, a legacy .NET upgrade, two of our own AI products and our own CRM. AI did the volume. The judgement, sign-offs and client relationships stayed with us. ### Can a two-person team handle enterprise clients? Yes. Several of the projects above were for global brands, delivered alongside their agencies. ### Who writes the code? AI writes the first pass. We design it, review it, test it and own it. ### How do you keep client data safe with AI tools? Client data stays inside approved tools and accounts, and on-premises builds are an option for data-sensitive work. ### What if a project needs more people? We'll tell you at the start, before anything's signed. ## It's Q4. Got a project that needs to get DONE before 31 December? - [Email us](mailto:hello@hiaitus.ai?subject=Q4%20project%20(two-person%20article))