David Johnson-Igra Lost All His Ghostwriting Clients to AI in 2025 — Then Rebuilt His Business Selling AI-Powered Executive Voice Systems
A tech ghostwriter who worked with Amazon, a16z, Meta, GitHub, and OpenAI watched his client roster disappear after Claude 3 Opus launched. He rebuilt using Claude, Obsidian, and MCPs — and now sells the knowledge infrastructure, not just the writing.
Continue exploring this workflow
The Strategy
David Johnson-Igra spent years crafting the public voices of tech executives at companies including Amazon, a16z, Meta, GitHub, and OpenAI. In April 2025, within weeks of Anthropic releasing Claude 3 Opus, he lost essentially his entire client base. The same executives who had hired him to write their LinkedIn posts and thought leadership content had either started using AI directly or cut content budgets in anticipation of doing so. Rather than compete with AI on price, Johnson-Igra decided to compete on context depth. His new service is built on a specific observation: the reason AI-generated content fails for most executives is not writing quality — it is that the model has no idea who the executive actually is. It does not know their opinions, their history, their audience's trust triggers, or the positions they have publicly taken. He builds that context layer and sells it. His system uses Obsidian as a knowledge graph to store every interview, post, article, and public statement an executive has made, tagged by topic and tone. Python scripts analyze historical LinkedIn engagement data — sometimes 4,000 rows or more — to surface which topics and formats have resonated with that executive's specific audience. A competitor audit of up to 1,000 content pieces across peer executives maps the positioning landscape. Claude then generates content drafts that draw on this context via MCPs, producing output that reads in the executive's documented voice rather than a generic AI voice. He describes the shift as moving from selling outputs to selling systems: "Now those outputs are a means to an end. The end is the system."
How It Works
Build a knowledge graph of the executive in Obsidian: compile every published interview, article, LinkedIn post, and public statement, tagged by topic, tone, and the context in which it was made.
Run a Python script against the executive's full LinkedIn post history to analyze engagement rates, impression counts, and audience response patterns. Identify which topics, formats, and angles have historically performed best with their specific followers.
Conduct a competitor content audit: pull and analyze a defined set of peer executive profiles to map over-served angles and identify positioning gaps the client can own.
Configure Claude via MCP to query the Obsidian knowledge graph as a source when drafting new content. This gives the model access to the executive's documented positions, past quotes, and voice characteristics at generation time.
For each new content request, provide Claude with a brief — topic, platform, goal, and tone notes — and generate a draft informed by the full knowledge graph context rather than from a cold start.
The executive's review process shrinks to approving a draft that already sounds like them and references positions they have actually taken, rather than correcting generic AI output or writing from scratch.
Results
Johnson-Igra does not disclose current client count or revenue figures in his Fortune coverage. Documented project specifics include a Python analysis of 4,000 rows of LinkedIn data and a competitor content audit of 1,000 pieces across 10 executives. He lost his previous client base in April 2025 following the release of Claude 3 Opus. These details are as reported by Fortune in May 2026.
Our Take
This is the correct response to AI eating a professional service: move up the value chain to the thing AI cannot do alone. The writing itself is increasingly commoditized. The context architecture — knowing exactly what an executive has said before, how their audience responds, and where they sit competitively — is not. That context takes research, judgment, and ongoing curation that a language model cannot provide without a human building and maintaining the underlying knowledge base. The Obsidian choice is deliberate and worth copying: it is file-based, model-agnostic, and does not lock you into any particular AI provider's knowledge format. As the model landscape shifts, the knowledge graph stays portable. The pattern here is replicable beyond ghostwriting. Any professional services firm that helps clients communicate — PR, investor relations, political communications, executive coaching — can build a similar context-depth advantage by investing in knowledge infrastructure that AI tools alone cannot replicate.
Frequently Asked Questions
The practical questions a builder or operator is likely to ask before trying a strategy like this.
What does this agency AI agent actually do?
This agency AI agent is a real workflow where the agent takes on an operational job, not just a brainstorming task. David Johnson-Igra Lost All His Ghostwriting Clients to AI in 2025 — Then Rebuilt His Business Selling AI-Powered Executive Voice Systems shows what that looks like in practice. A tech ghostwriter who worked with Amazon, a16z, Meta, GitHub, and OpenAI watched his client roster disappear after Claude 3 Opus launched. He rebuilt using Claude, Obsidian, and MCPs — and now sells the knowledge infrastructure, not just the writing. The practical value comes from the agent handling repeatable business work with enough autonomy that a human only steps in after context has already been gathered.
Who should use a agency AI agent like this?
This example is most relevant for agency operators. It is especially relevant for businesses where speed to lead, after-hours coverage, or consistent intake quality directly affects revenue. The category here is Content Creation, which means the best fit is a team looking to turn a manual bottleneck into a repeatable system with a agency AI agent.
Which tools are used in this agency AI agent setup?
The source names Claude, Obsidian, Python. That matters because one of the strongest signals in this directory is whether the operator shared the actual stack. Named tools make a agency AI agent strategy far more useful than vague claims about “an AI system” doing the work.
How hard is it to implement a agency AI agent like this?
Advanced difficulty is the current read. The listing suggests a launch window of weeks. Startup cost is listed as $50-200/mo. We were able to extract 6 concrete workflow steps from the source. We would treat a agency AI agent like this as a workflow that needs real business context, testing, and exception handling rather than something you should copy blindly from one prompt.
What results can a agency AI agent produce?
Johnson-Igra does not disclose current client count or revenue figures in his Fortune coverage. Documented project specifics include a Python analysis of 4,000 rows of LinkedIn data and a competitor content audit of 1,000 pieces across 10 executives. He lost his previous client base in April 2025 following the release of Claude 3 Opus. These details are as reported by Fortune in May 2026.
How credible is this agency AI agent case study?
Right now the evidence comes from an article from fortune.com. That is enough for us to study and curate the workflow, but not enough on its own to treat this agency AI agent like an audited case study. We look for named tools, concrete results, and enough workflow detail to understand what was actually deployed, then we add our own editorial judgment on top.
Related Strategies
More AI agent strategies you might find useful
52 AI Generated YouTube Videos in 6 Weeks With 30,000 Views and Zero Manual Editing
An autonomous AI pipeline published 52 YouTube videos in 6 weeks, generating 30,…
Nate Herk Built an n8n Agent That Finds Viral YouTube Patterns and Earns Him $6K Per Month
A no-code n8n workflow that scans any YouTube niche for viral outliers, extracts…
Nate Herk Built an AI Voice Agent That Calls Every New Lead Within Seconds of Their Form Submission
An n8n + Vapi workflow that triggers an outbound AI phone call the moment a lead…
A Thoracic Surgeon Handed His YouTube Channel to Two AI Agents for Six Weeks. Here Is What Actually Happened.
Wei-Ciao Wu built two Claude agents — Midnight for production and Dusk for promo…
Want more strategies like this?
Get weekly AI agent case studies in your inbox.