About
I’m Craig. I help companies put agents to work across the whole business.
I’ve spent more than 20 years building products and leading teams. These days I work with companies from 5 people to 500, helping them use agents in sales, operations, client delivery, marketing, finance, support, and software. The tools get more capable every week. The hard part is figuring out where they actually belong and making the change stick.
Why I started Vibecto.
I’ve been building with these tools since GitHub Copilot first showed up, but the work didn’t stay in the codebase for long. Now I use agents for call prep, follow-up, bookkeeping, CRM updates, research, writing, internal operations, and yes, shipping software. I’ve learned where they create real leverage and where they just create more mess. Eventually, people I respect started asking some version of the same question: can you come do this with us?
That’s Vibecto. I work with a small number of companies at a time because I want to stay hands-on. The company can be five people or five hundred. We start with how the work actually gets done, find where agents can take something on or make something newly possible, build the first useful systems, and help the team keep going. No giant transformation theater. Just useful things running in the business.
What I’m using right now.
This changes constantly, so I put dates on it. These are tools I’ve actually used enough to have an opinion about, not a list of every product I’ve tried for an afternoon.
Last reviewed
Agents I work in
- ChatGPT Work / CodexAug 2025–present
This has been my daily driver since June 2026. I use it for sales prep, follow-up, research, writing, internal operations, recurring automations, and software delivery—often with several agents working in parallel.
- Claude CodeMay 2025–present
I still use Claude Code for scoping, exploring unfamiliar codebases, and getting a second opinion on work I did in Codex. I like having the robots check each other’s work.
- Claude CoworkJan 2026–present
I bounce between Cowork and ChatGPT Work for general knowledge work. I still prefer Claude for a lot of coaching routines because the responses feel more human. Using it regularly also means I can help Cowork-first users without bluffing, and I’ll know if it gets good enough to pull me back in full time.
The operating layer
- Skills, sub-agents, and automationsNov 2025–present
When I do something more than once, whether it’s software delivery, call prep, bookkeeping follow-up, or content work, I try to turn it into a reusable skill or automation. I also hand bounded or context-heavy work to sub-agents so I can keep moving without stuffing everything into one giant conversation.
- MCPs and purpose-built CLIsJul 2025–present
These are how I let agents work in the tools I already use: email, Slack, calendars, browsers, databases, internal systems, GitHub, and Linear. I usually prefer a purpose-built CLI when there is one and MCP when it’s the better fit.
- GitHubFeb 2012–present
GitHub is still the center of how I ship software: repositories, issues, pull requests, collaboration, and a durable history of what changed.
- LinearMar 2025–present
This is where I plan and track my own work and a lot of client product work. I’ve wired it into the agent workflow from creating a ticket all the way through shipping the pull request.
- Ace + ObsidianDec 2025–present
Ace started as my home-grown AI executive assistant. Now it’s partly a set of agent-native tools I use from Codex and Claude, and partly a projects-and-goals system stored as Markdown in Obsidian because that fits my brain better than another database.
- MonologueSep 2025–present
I started using Monologue for dictation shortly after it launched, replacing Wispr Flow. On August 20, 2026, it also replaced Granola for meeting transcripts, which made that context much easier to hand to agents through an MCP, API, or CLI.
Models and runtimes
- OpenAI and Anthropic modelsOpenAI: 2022–present; Anthropic: 2023–present
These are still my defaults. I pick between the models and the products wrapped around them based on the job instead of pretending one company is going to win every category forever.
- Open-weight and local modelsDec 2025–present
I test Qwen, Kimi, DeepSeek, and other open-weight models through Ollama, vLLM, and hosted services. I use them when cost, privacy, control, speed, or a specific capability makes the tradeoff worth it. They aren’t my default yet.
- Next.js, TypeScript, Neon, and VercelNext.js: Jul 2022–present; current stack: Dec 2025–present
This is a common stack in my current hands-on product work. It’s a preference, not a religion—I’ll use the language, framework, database, or platform that fits what we’re building.
Things that fell out of the rotation.
None of these are necessarily bad tools. I just don’t reach for them by default anymore. Why I stopped using something is often as useful as why I picked it up.
- GitHub CopilotJun 2022–Jan 2025
This was my main AI coding assistant from its public launch until Cursor’s Composer workflow made it feel like I could hand off a whole task instead of asking for the next few lines.
- CursorSep 2024–May 2025; daily driver by Jan 2025
I started playing with Cursor in September 2024. By January, Composer and agent mode had become how I did most development work. Then Claude Code pulled me into the terminal.
- GranolaJan 2025–Aug 2026
Granola captured my meetings for more than a year, and I fed the transcripts into follow-up and agent workflows. I moved that work to Monologue after my last Granola transcript on August 19, 2026.
- Wispr FlowJun 2025–Sep 2025
I used Wispr Flow for dictation starting June 23, 2025, then replaced it with Monologue shortly after Monologue launched that September.
- SupabaseJul 2025–Dec 2025
I used Supabase for the Vibecto site, then moved it to Neon and Drizzle. I’d still use Supabase when it fits the product; it just stopped being the right choice here.
- n8nOct 2025–Apr 2026
I used n8n for precise, editable visual workflows, including paid client work. Most of that work has moved into simpler Codex and Claude automations, but n8n is still useful when a visual canvas and fine-grained control actually matter.
- OpenClawFeb 2026–May 2026
I used OpenClaw to run persistent personal and business agents through Slack. I stopped when the Claude API bill felt too high for the value I was getting, and the cheaper models weren’t reliable enough. Most of that work now runs as automations in ChatGPT Work and Claude.
A few things I’ve made or written.
Where I’m based.
I’m in Indianapolis. I mostly work remotely with companies across the Midwest and beyond, with some travel when it’s useful.