Building With Claude Since 2023: What Holistic Consulting Technologies Has Actually Shipped

HCT has been building with Anthropic's Claude since late 2023 — practically from the time it became publicly available. What started as voice recordings and manual prompt engineering has evolved through three distinct eras: the prompt engineering phase, the context engineering phase enabled by Claude Projects, and now the Claude Code phase of agentic software development. Each era produced different capabilities. All of it built toward the same goal: making AI deliver real, measurable results for Charlotte and Lake Norman small businesses.

How We Started: Voice, Prompts, and Google Docs

When Claude first became available, we did not have agentic workflows or structured pipelines. What we had was a process that felt manual by today's standards but was already producing results no traditional consulting workflow could match. We would record client sessions, run voice-to-text transcription on the conversation, and then manually prompt-engineer the research, strategy, and deliverables through the chat interface — working from a curated set of prompts we had refined and organized in Google Docs. Each engagement was a deliberate sequence: surface the problem in the transcript, feed context into Claude, iterate on the output, refine the strategy. It was methodical, it worked, and it taught us exactly how Claude reasons and where the leverage points are.

That early phase built something you cannot buy: an intuition for what Claude actually does with context, where it drifts without constraints, and what kinds of framing produce consistent strategic-quality output versus generic text. The Google Docs prompt library was essentially our first knowledge base — manual, but purposeful.

The Context Engineering Era: Claude Projects

When Anthropic released Projects, everything accelerated. Projects gave us persistent context, a structured knowledge base, and the ability to configure Claude's behavior through project-level instructions and documents. We stopped thinking in individual prompts and started thinking in context architecture — what does Claude need to know about this client, this engagement, this business domain, before the first question is even asked? We moved our best prompt frameworks into project documents, uploaded relevant client materials into the knowledge base, and tuned the project settings to enforce consistency in tone, format, and strategic framing across every interaction. The result was a step-change in output quality. The same strategic reasoning we had been producing manually became reproducible, faster, and scalable across multiple engagements running in parallel.

Context engineering — designing what Claude knows before it starts working — turned out to be the highest-leverage skill in the entire workflow. It is not about writing better prompts. It is about building the environment in which Claude operates. That insight is what separates clients who get consistent value from Claude from those who get inconsistent results despite using the same tool.

The Claude Code Era: Agentic Builds and Real Software

When Claude Code launched publicly in 2025, we were among the first to use it for production work. The first major project: a private K-8 school came to us with a problem most software tools cannot touch — years of custom-developed rubrics and standards built on top of North Carolina state standards, differentiated across every grade level, living in large folders of loosely organized documents. Every teacher knew the material. Nobody could query it, cross-reference it, or feed it to a platform. We used Claude Code to build a transformation pipeline that ingested the full document set, applied consistent naming conventions, deduplicated overlapping standards, flagged and corrected errors, and produced structured output files alongside ready-to-run SQL — making database creation a matter of clicks rather than weeks of manual work. That database became the curriculum backbone of BranchBase, our custom-built AI educational platform for the school and the direct predecessor to HomesCool.

What We Have Actually Built

The K-8 curriculum project was the first, not the last. We have used Claude since late 2023 for strategy, research, automation, and training work — and Claude Code since 2025 for software development and agentic builds. Not as productivity add-ons, but as the foundation of how we design, build, and deliver. The project types span the full range of what small businesses actually need.

For business owners who needed software built to their exact operations, we used Claude Code and rapid prototyping methods to ship custom applications on timelines traditional development cannot match. GitHub's controlled study of 95 professional developers found those using AI coding assistance completed tasks 55% faster than those working without it — a productivity delta we have experienced directly across dozens of engagements in Davidson and Charlotte.

For clients who wanted their research, strategy, planning, and delivery processes automated end-to-end, we built agentic consulting workflows: systems that take a business challenge as input, run structured research, generate strategic recommendations, produce a delivery plan, and output formatted client-ready documents — without manual intervention at each stage. This is Claude used as an orchestrator, not just a writing assistant.

The data transformation work — converting large volumes of unstructured documents into structured, queryable databases with automatic naming conventions, deduplication, and error correction — has become one of our most requested services. Most organizations are sitting on years of knowledge trapped in files that cannot be searched, compared, or used by software. We unlock that data systematically, using Claude Code at a scale and speed no human team could replicate.

Training Teams to Work With Claude Effectively

A significant portion of our work is not building — it is teaching. Business owners and their teams who want to use Claude themselves need more than a prompt-writing tutorial. They need to understand context management, output consistency, and how to design agentic workflows where Claude takes sequential action without constant human input.

We deliver this AI training in-person and virtually, from beginner sessions — what is Claude, how does it reason, why does context matter — to advanced workshops on Claude Code, computer use, and multi-step automation. Engagements run from one focused week to a full 90-day transformation depending on the scope.

62% of small businesses that adopted at least one AI tool reported significant productivity improvement within six months. The companies in that 62% almost universally had someone who did structured implementation work first. The ones outside it did not.

Why Access Is Not the Advantage

Every Charlotte and Lake Norman business can create a Claude account today. Access is not the differentiator. Most automation initiatives fail to deliver the ROI initially expected — meaning most organizations with the same tools do not achieve meaningful results from them.

The gap is applying AI to your specific workflows, your specific data, and your specific business logic. That knowledge comes from shipping things, making mistakes in production, and learning where Claude creates compounding value and where it needs human judgment to stay on track.

AI investments return an average of $3.70 per dollar invested, with top performers reporting $10 or more per dollar invested. The difference between those outcomes is not which tool was used. It is how well the implementation was designed.

The businesses that get real results from Claude are not the ones who have access to it. Everyone has access. The ones who get results are working with someone who has built with it, shipped with it, and knows exactly where it creates leverage.

Over two years with Claude — and nearly a year with Claude Code — produces something reading about AI cannot replicate: a systematic understanding of what these tools can own in a real business workflow, and where they still need human judgment. The questions we know how to answer — because we have answered them in production — are the questions most Charlotte businesses are still trying to figure out alone.

If you are ready to move from experimenting with AI to actually deploying it, start with a conversation. Reach out to the HCT team →