Guidable.
Paul Wakim.
I've been building businesses since graduating college in Pittsburgh in 2016. What started with door knocking and cold calling became one of the largest real estate media companies in Pennsylvania.
Through that business, I dove into real estate investing and had an idea: what if I could predict who was going to sell their house next? I tried teaching myself to code to build that algorithm. I failed. But I succeeded in finding a co-founder who could actually build it.
In late 2019 we launched a company that used predictive algorithms to identify houses likely to sell before they hit the market. That business ignited my obsession with technology, automation, and artificial intelligence. When ChatGPT launched, we were among the first to integrate it into our operations. We sold that company in 2024, and since then, I've been all-in on AI and automation.
Now I'm consulting because there's nothing I enjoy more than helping people discover what's possible with AI.
Three services.
Hands-on Claude training for executives, ending with a working personal AI assistant for every participant.
Desktop, mobile, and web applications, built new or finished from where you got stuck.
AI agents engineered to run a business process from start to finish on their own.
Things I have built.
- PodSips: chat with all podcasts.
- Tiny Family Tales: personalized story books for families.
- DAM Nutrition: a calorie and nutrition tracking app.
- 15 Layers: fully autonomous AI supply chain tracking and investment analysis.
Book a call.
Hands-on Claude training for executives.
Training for groups of two to five executives, or one-on-one, that starts at the standard chat window and ends at the most advanced features of Claude Code.
The outcome: every participant leaves with a working personal AI assistant connected to their real email, calendar, and files.
Three levels. Each builds on the last.
- How Claude works: what the model is doing and why it sometimes gets things wrong.
- Prompting: writing instructions that get good results.
- Projects: workspaces that hold standing instructions and reference documents.
- Connectors: links that let Claude work from your real email, calendar, files, and business tools.
- Memory: Claude remembering you and your work between conversations.
- Research mode: Claude investigating a question and returning a report with sources.
- Documents: Claude creating real Excel, Word, PowerPoint, and PDF files.
- Office add-ins: Claude working inside Excel and PowerPoint themselves.
- Claude in Chrome: Claude operating the browser to complete tasks on websites.
- Chat, Cowork, and Claude Code: the three ways to use Claude, and when to use each.
- The three interfaces: running Claude Code in a plain terminal, the desktop app, or a code editor.
- Working in a folder: Claude reading instruction files up front, then other files as each task requires them.
- CLAUDE.md files: standing instructions at the personal, project, and sub-folder level.
- Skills: repeatable procedures stored in folders, with reference documents and scripts.
- Plugins and MCPs: connections between Claude and your other tools.
- Permissions: control over what Claude may do without asking.
- Checkpoints and rewind: undoing anything Claude did.
- Plan mode: Claude writing a plan for your approval before changing anything.
- The assistant folder: a dedicated folder with instruction files and persistent memory.
- Tool connections: linking the assistant to the tools you run your business on.
- Personal skills: two or three procedures built for your own recurring work.
- /goal: Claude working until a stated outcome is met.
- /loop: Claude repeating a task on a timer while the session is open.
- Routines: scheduled agents that run on Anthropic's servers while your computer is off.
- Remote access: driving the assistant from your phone while it runs at home.
The optional advanced track.
For the people who finish the three levels and want to build systems, this track runs as its own engagement.
- Sub-agents: specialized workers with their own instructions and permissions.
- Dynamic workflows: a script that directs many agents at once on one large job.
- Hooks: automations the software runs at set moments, every time.
- Parallel work: several sessions at once, and teams of agents that message each other.
- Harness building: combining these pieces into a system built for one business process.
Loops, routines, managed agents.
- A loop repeats a task inside an open session on your machine.
- A routine runs on Anthropic's servers on a schedule you set, with your computer off.
- A managed agent is built by a developer into their own software product.
Paced to the group.
Each level takes as many sessions of two to three hours as the group needs, and most groups need more than one session per level.
Guidable issues a certificate of completion at the end.
Working software, delivered running.
Guidable builds desktop, mobile, and web applications, and improves applications you already have.
You receive the application live and hosted, with the source code and documentation.
Build new, or finish what exists.
- Build something new: you describe the product or internal tool, and Guidable designs it, builds it, and delivers it running.
- Finish or fix something that exists: Guidable repairs the bugs, adds the missing features, and cleans up the code.
Many owners now get an application most of the way done with AI tools, and Guidable finishes the remaining work: payments, user accounts, bugs that appear with real data, and hosting.
The deliverables.
- A working application.
- The full source code, owned by you.
- Hosting set up and documented.
- A written explanation of how the application is put together.
Scope. Build. Deliver.
- Discovery: conversations that produce a written scope with the exact features, the price, and the delivery date.
- Build: progress shared at agreed checkpoints, so you see the real product early.
- Delivery: the application goes live, you receive the code and documentation, and an agreed period covers fixes afterward.
Agent and harness engineering.
Guidable engineers AI agents: systems where an AI model carries out a business process from start to finish in defined steps, with little or no person involved while it runs.
An agent is a model plus a harness, and the harness is the software built around the model that runs its tools, keeps its memory, checks its work, and controls what it may do.
The harness decides how well the agent works.
The same AI model performs well or poorly depending on the harness around it, so the harness is where Guidable puts the engineering effort.
- Tool design: the exact set of actions the agent may take, each with its own checks and permissions.
- Instructions and context: control over what information the model sees at every step.
- Memory and state: records of finished work, so the agent continues correctly across steps and sessions.
- Guardrails: fixed rules that block certain actions no matter what the model decides.
- Evaluations: a standing set of tests that every change must pass before it ships.
- Observability: a complete record of every step, tool call, and decision the agent made.
- Human checkpoints: a person approves the expensive or irreversible steps.
- Error recovery: the agent verifies its own output, retries failed work, and rolls back mistakes.
- Orchestration: many agents working in parallel, with defined handoffs between them.
When a process qualifies.
A process is a good fit when all three are true:
- The same kind of work arrives over and over.
- A knowledgeable person could write down the steps.
- Each step needs judgment (reading, drafting, classifying, or comparing) that today takes a person's time.
Design. Build. Run.
- Discovery: Guidable studies the process as you run it today and produces a written system design with a fixed price.
- Build: the pipeline is built step by step on your real material, with outputs reviewed together.
- Run and hand over: the system runs on Guidable's side as a service, or is installed on yours with your team trained.