The Challenge
Mark Rhodes and Steve Jacobs met in graduate school at Harvard in the early 1980s. They have spent their careers in organizational strategy and design consulting, often working together.
With Strategy for the AI Era, they wanted to make that expertise available to small business owners who needed to work through a strategic question on their own.
Their newest project, Risk Management: Where AI Leaves You Exposed, focuses on four questions: Does the business understand its insurance coverage for AI? How is it protecting its intellectual property? What is the team putting into AI tools? And could outside content trick those tools into following unwanted instructions?
The goal was a practical assessment that ended with clear next steps. An owner should know which questions they could work through themselves and which ones belonged with a broker or attorney.
Building that experience took more than writing a single prompt. Their broader suite produces reports of up to ten pages, including tables, charts and strategy models tailored to the business. The tools needed to retain business context, finish lengthy outputs and keep search costs under control.
The Solution
Mark and Steve built their tools on Pickaxe and embedded them on their own website.
The risk assessment now has four linked stages, beginning with insurance and continuing through IP protection, team AI use and protection against malicious instructions, including prompt injection. Each stage links to the next, giving owners a starting point and a path through the assessment.
Owners answer questions about what they already have in place. If they do not know, they can say so. The tools assess the answers, identify areas to investigate and suggest specific questions to take to the relevant professional.
The work begins before the agent does. Mark and Steve write out a detailed method for each tool: what it should ask, what it should check and how the report should read. They refine that method through testing before turning it into a Pickaxe prompt.
They also include a library of named sources in the prompts to ground the analysis. The intended output is plain language tied to the owner's circumstances, with a clear distinction between practical follow-up and questions that require professional judgment.
Memory connects the tools
Pickaxe User Memory carries business context from one tool to another, helping the suite build on earlier conversations.
"No platform we've tried holds on to what it learns about a user the way Pickaxe does. That's what lets a set of separate tools feel like one advisor who knows your business."
For findings they want to preserve exactly, they built a custom Action that writes the finding to a named memory trait and returns a receipt. The agent tells the user a finding has been saved only after receiving that confirmation.
Settings match the real work
Their reports exposed a practical constraint: default settings did not always fit the longest output.
Mark and Steve tuned input length, output length and the amount of memory each agent reads against their longest real report. Mark says that adjustment stopped long reports from cutting off partway through.
They also narrowed the search tools available through custom Actions. That let each agent access the basic searches its job needed without opening up more expensive research or crawling operations.
When one risk agent's prompt became too large to run reliably, they separated the conversation and report-writing work into two agents on the same page. That is a build choice within the assessment, separate from the four topics owners work through.
What They Built
The result is a connected set of tools that turns a business owner's answers into an assessment and an ordered action list, including who should handle each step.
The concrete achievement is the working product: consulting methods translated into guided conversations, business context carried between tools and substantial reports delivered through their own website.
Mark's build lessons are equally useful to other creators:
- Test the longest output. Set lengths against a real report, including its tables and charts.
- Confirm important writes. Have an Action return a receipt before telling the user a finding is saved.
- Limit tool access to the job. Keep search capabilities aligned with the agent's purpose.
- Separate work when a prompt gets too large. Conversation and report generation can have different responsibilities.
- Read settings back. After making changes through Wingman, ask it to return the stored value so you can check it.
- Use your own business as a test case. Mark and Steve tested the assessment on themselves before launch.
The Takeaway
Mark and Steve brought the consulting method. Pickaxe gave them a way to turn it into tools owners could use on their own website, with memory connecting the work across sessions and agents.
Their approach offers a useful starting point for consultants: write down the method, build a guided version, test it against the hardest real output and connect the parts that need shared context.
Explore Where AI Leaves You Exposed. Mark describes the launch experience as free with an email address. The assessment helps owners prepare their next steps and questions for a broker or attorney; it does not determine insurance coverage or provide legal advice.