Ethos

This is what I actually care about and what I’m focused on solving.

The models aren’t the problem

Companies spent millions trying to bring AI in — software licenses that promise to take work off the plate, or straight to model providers in token spend. Leadership buzzes about going AI-first. Ask what changed day-to-day and the answer is some version of nothing.

The models are good enough. Stop blaming the models. Enterprise failure rates have stayed flat across model generations. Process is the bottleneck.

Intelligence is commoditized. Advantage is deployment — where, how, and why it enters a specific company.

That’s the FDE job. I do this as a Forward Deployed Engineer at ninjacat.io: harness intelligence to company-specific context, not bolt on another generic copilot.

Why engineers got the gains

AI is working at scale for one group right now: software engineers. Their work is bounded, checkable, structured, and verifiable. Compilers, tests, diffs, version control. Feedback in seconds.

Most enterprise functions aren’t like that. Finance close, sales ops, marketing, AP — the real process lives across ERPs, CRMs, Slack threads, spreadsheets, and unwritten rules in people’s heads. Point a capable model at that without redesign and you often get negative ROI: the same work, plus time correcting the AI.

That same wave is opening the gate. Anyone can be a coder now — everyone should be a builder. What’s scarce isn’t syntax. It’s creativity, communication, distribution, and the ability to sell. Ideas don’t have to wait behind a specialist queue anymore.

How deployments fail

• Skip the audit. The documented process is rarely the real one. Build for the SOP and you automate 70% of the volume and break on the 30% that creates more work than before.
• Throw everything at the LLM. Tokenmaxxing. Production systems that work are mostly deterministic code, with the model only where judgment actually lives.
• Agent sprawl. Everyone vibe-codes their own agent. No shared spine, no governance, no one on call when something breaks.
• Ship and move on. AI is infrastructure that keeps shifting underneath you — not a project you declare done at go-live.
• Force migrations. If the pitch is “leave the system you spent years adopting,” the transformation stalls. Build on top of what already runs the business.

What works

• Audit before build. Sit with the people doing the work. Map exceptions. Decide what not to automate.
• Decompose the workflow. Deterministic software where rules and inputs are predictable. Agents where the object is clear but the path varies. Humans where the decision carries ambiguity, accountability, or irreversible consequences.
• Build where the work already happens. On top of existing systems of record. Outcomes, not another interface employees have to adopt.
• Evals and human-in-the-loop before wider autonomy. Prove the system behaves; feed corrections back in.
• Loop, don’t one-shot: Audit → Build → Evals → Deploy → Observe → Improve.

Human-first

AI will replace tasks. Most jobs are those tasks. How we structure the future matters.

Lead with empathy, compassion, and care. Bring people on the journey. Keep humans in control of the important parts of the loop — approval, accountability, the calls that shouldn’t be silent. Change lands differently when it touches someone’s livelihood. Sell the future honestly. Never ask someone to disappear into the model.

Trust is earned face-to-face and in how the system behaves when it fails. Widen what a person can be. Keep care in the loop.

North star

Stop waiting on the next model release. Redesign the work to meet the intelligence already here.