When I left my last employer in March 2025 after six years, I found MADE, Inc. sitting exactly where I'd left it in 2019: a quick & easy WordPress site that looked like the abandoned brochureware that it was. The job market has been, well… interesting. Slow interesting. (Mostly slow.) Maybe I'll find that perfect next role while I do volunteer work, or maybe it will make more sense to consult again. In the mean time, I had an opportunity to show what MADE, Inc. could be, and what results I can deliver.
I'd already been rebuilding my personal brand ecosystem. I (finally!) got brianfending.com back after losing the domain for a decade (a long, sad domain squatting story). Not satisfied, I built ai.brianfending.com as a sort of “inbound screener” with a RAG-trained bot with access to my resume, teeing it up as something helpful for my (then-new) job search process to help hirers/recruiters learn if I've worked with specific problems, technologies, or frameworks over the years.
But MADE, Inc. needed something different. Personal branding is about thought leadership and writing, but a consulting business needs to demonstrate methodology, not just talk about it.
The Archaeology
The old site was a time capsule from a different era of my career, with set-and-forget pages describing technology governance and product development consulting in appropriately vague consultant-speak. A few blog posts I'd written when interesting problems came up at work. It worked well for the time, but functionally did not age well. The content was evergreen, for sure, but telling people about my expertise proved nothing.
The fundamental problem with most consulting websites is they describe what you can do without demonstrating how you think and execute. Anyone can claim expertise in AI governance or product development. (See also: 98% of LinkedIn…) But I wanted something that would make prospects experience a process of my design before they ever talked to me. Just going through a short process of introspection should help potential customers understand the depth of thinking involved and it may open their eyes to things that wouldn’t come up organically during a discovery meeting.
Dev Environment: AI-First Workflow
The rebuild happened entirely with Claude Code in VS Code. As people who’ve read other posts of mine know, I take my approach to AI-assisted coding very seriously, and having it directly in the development environment changes how you approach complex projects, especially when you're working alone and need to maintain context across multiple domains.
I organized everything in a single VS Code workspace with four distinct directories:
madeinc-nextjs: Main application repository with standard Next.js structure of components, pages, API routes, and all the TypeScript awesomeness you’d expect.
madeinc-context: Eight context documents that became the project's knowledge base. These weren't just notes, but structured topic guides for AI agents to understand business requirements, assessment methodologies, and implementation decisions. A “PRD on steroids” one might say.
madeinc-wordpress-code: Extracted theme files, plugins, and configuration from the old WordPress site. Useful for understanding how content was structured and what functionality needed to be replicated or improved. Mostly for theme mods but by all means, let’s save those four blog posts, too.
madeinc-wordpress-data: All the old content exported to Parquet files from the S3 bucket they were rotting in using Python. Parquet handles the mixed data types (text, metadata, timestamps) better than CSV and made it easy to analyze content patterns when designing the new information architecture.
Visual Design
I tried remembering 1% of my image editing skills but I am slooooow in an image editor. So I used ChatGPT-accessed image models to revive and modernize my logo and signature art from the original site. Honestly, I made some really big improvements before taking everything to Canva for a tweak or two. It surely helped that I had existing assets that needed a refresh instead of a blank canvas. Much faster than relearning layer masks and vector paths.
Context Context Context
I’ve written about this this year, but context in the developer experience is changing not only team toplogies, but how even solo devs can work more effectively by mimicking that larger team. The madeinc-context directory contained the project's institutional memory. For starters:
Technical Architecture Decisions,
Content Migration Strategy, and
Performance Analytics.
From there, I started getting creative with the idea of a set of assessments. For that, I had additional docs for:
Assessment Methodology,
Scoring Logic,
UX Requirements,
AI Integration Patterns, and
Business Process Flows.
Each document was written to train AI agents on project-specific knowledge that was unique to my project, and then referenced from and quoted in the madeinc-nextjs project’s CLAUDE.md and README.md files.
Agents Agents Agents
Meanwhile, agents (try /agents in Claude Code if you haven't already!) handled syncing documentation and providing test coverage as the project evolved, while I reviewed and pushed git commits manually into feature branches for each major component.
Honestly, this was pretty clutch and something I’ll do with every project from now on.
Building Assessments That Actually Matter
You can't just prompt an LLM to "generate 20 questions about AI governance" and expect useful results. The questions need to reveal actual business problems, not surface-level checkboxes akin to the most basic of compliance management platforms.
The assessment questions needed to be based on benchmarks for which there are data, derived from sources that I had read and respect. Once I had a good sense of what each paper or study measured, the questions evolved pretty naturally. These will of course need to change over time as the market evolves, along with updated sources upon which to base new assessment questions. I started out just looking at AI because that’s a hot topic, but wound up with four assessment tracks based on areas where I'd consistently seen organizations struggle, and in which I’ve worked deeply over the past decade in particular:
AI Risk Assessment: Most companies are running 3-5x more AI applications than they realize. I lived that in a small (35-seat) org when ChatGPT 3 first reared its head, and I have to say it’s not getting any better thanks to general proliferation of AI among staff and AI-related feature creep in SaaS platforms already embedded in organizations.
Product Development Maturity: A lot of organizations succeed at product development in spite of their own lack of process. That will always mystify me, but the gap between having a process and having a scaling process is vast. Teams that work fine at 10 people often start to collapse at 25, and get a reorg at 50.
Technology Governance: The difference between governance theater (elaborate frameworks nobody follows) and governance that actually enables faster, safer decisions. Often conflated with Compliance, which is a rant I documented recently on LinkedIn.
Security Posture Assessment: I am sure your checklist is awesome, and maybe your incident response plan won an award. But Posture isn’t just what you do when somebody’s looking, rather it’s best demonstrated by anticipating patterns and adapting to circumstance before an incident.
Each assessment needed questions that distinguished between someone who talks about these domains and someone who's actually implemented solutions. The AI risk questions, for instance, probe whether you understand the difference between AI usage oversight and AI development governance.
Scoring methodologies can be hard, so I tried to keep it simple yet fair. The question groupings are key as a rollup, because individual responses matter less than patterns across categories that reveal a truer organizational maturity level.
From Logic Trees to Smart(er) Prompts
I started building the assessment engine using procedural logic. If someone scores high on data governance but low on AI oversight, recommend X. If they have formal processes but poor adoption, suggest Y.
This seemed right, and for the quantitative aspects it worked well. I could control every variable, account for every combination, ensure consistent recommendations. As I started to apply that to the executive summary, though, I started to lose nuance. And the result of those nested conditionals read like a garbage McKinsey report circa 2015. Consultant Mad Libs!
So after careful consideration, I bit the bullet and pivoted to using AI for just the executive summaries. I fed Claude the assessment data along with A LOT of context about what these scores actually mean in practice. Honestly, I didn’t iterate a lot after spelling out everything in the instructions. It was better and more succinct than I could have hoped.
Claude understood through its own training and my context that someone with high technical capability but low governance maturity faces different risks than someone with strong processes but poor technical implementation. It could generate insights about organizational dynamics and implications on change management that easily fell out of scope when I tried the algorithmic approach.
Quality Assurance: AI on AI Violence
When it came to review of the final product of one PDF summary per assessment, I created 16 assessment datasets (four per assessment type) and had ChatGPT 4o pitted against Claude Opus 4.1 in their roles as harsh, competitive critics. They assessed not just the AI-generated summaries, but the entire methodology and content. The models iterated their advice, and I reigned in the overly generic advice that assumed far more context about the client's environment than the assessment data actually provided. Only after repeating the exercise with many more datasets and manual review was I comfortable with the reliability of the content.
This became a sophisticated prompt engineering challenge. I needed consistency across assessment types while maintaining the intelligence that made AI-generated content valuable. Version control, testing against synthetic user data, and validation logic to catch responses that drift too far from established scoring models.
But there's a nuanced detail around the storage and versioning of the Claude-written executive summaries. When Claude generates content, my website’s API stores and versions it in the database, tied to the specific assessment and question set. If I ever change how those summaries are written with different prompts, updated methodology, new scoring logic – that versioning corresponds to what was actually stored and gives me options.
I can either keep historical assessments "frozen in time" with their original summaries, or regenerate them using updated prompts while maintaining data integrity. The versioning ties to the questions asked too, so as those evolve based on new research or changing industry standards, I can maintain consistency across different assessment generations AND leverage AI improvements.
Every report summary is tied to the exact prompt and scoring version that generated it, so recommendations are reproducible and auditable over time. This matters because consulting methodologies should evolve with new data and market conditions. Over time I'll be able to accrue data and aggregate it in meaningful ways, creating my own basis for reporting on industry trends, organizational maturity patterns, and implementation gaps across different sectors.
The platform can improve over time without breaking historical assessments or losing the ability to compare results across different periods. It's like having version control for business intelligence that also builds proprietary market research.
So ultimately, that AI integration made the platform more reliable, not less. My procedural logic had limitations, but well-engineered prompts consistently produced insights that align with how I'd actually approach these consulting engagements.
What Works
The platform now generates assessment reports that demonstrate methodology rather than just describing it. Prospects go through a diagnostic process, see how I think about their specific challenges, and get actionable recommendations they can actually implement.
"Step 0" introductions explain what each question reveals about their business before asking for an email address, setting up each assessment as its own landing page-style cell. In the process of considering the questions, people learn something about their own operations before I learn anything about their contact information.
The technical architecture scales consulting expertise without requiring consultant time for every interaction. The platform runs on Vercel's serverless infrastructure with GitHub integration for continuous deployment. When I merge changes to the production branch, GitHub Actions triggers deployment automatically, critical when iterating on assessment questions or prompt engineering.
The user can download their PDF directly or follow a link in a Postmark-generated email, and the system creates a work item directly into Jira. Every assessment becomes a qualified lead with detailed context about their maturity, challenges, and next steps. The conversion flow includes Calendly link at critical points, presenting immediate engagement opportunities that create a sales pipeline.
So in the end, the platform demonstrates capability instead of just claiming it. Whether I return to consulting or join another organization, this platform shows not just how IT professionals are working now, but how fast we can work when we use the right tools for the job.
Professional services are shifting toward a highly evolved "earn the business" approach for the AI age. When anybody can claim expertise, you need to demonstrate methodology and show results. Firms are giving away 1-2 hours of discovery work, spending a day or two prototyping, then delivering ready-to-implement solutions or compelling reasons to continue the engagement.
The old model was "trust me/us, I’m/we're experts." The new model is "experience our expertise, then decide." AI makes this approach scalable in ways that weren't possible before. You can automate the demonstration without losing the sophistication that makes consulting valuable. And I think this is exactly the kind of practical AI wisdom that organizations need right now.
CREDITS: cover image based on licensed artwork manipulated in Canva using app Pixeltone; Anthropic Claude for editorial review