Services/AI Tooling

Service · 01

AI Tooling & Consultation

Internal tools, agent workflows, and senior guidance for teams figuring out where AI actually earns its keep. Or — a paid discovery call to decide whether AI is the right spend at all.

Stack

Claude API · Anthropic SDK · Python / TS

Typical scope

1 internal tool, 1 agent workflow, or a discovery engagement

Timeline

2–6 weeks for a build · 1 week for a discovery

Claude · Agents · Internal Tools
“Where AI actually earns its keep.”
§ 01Overview

Most AI tools don’t survive contact with real work. These do.

I’ve shipped AI tooling for sales, marketing, customer success, and dev teams who’d rather use a focused workflow than a chatbot. The pattern is always the same — find the place where a human is doing rote reasoning, replace the rote part, keep the judgement call in front of a person.

NordDashboard is my own example of the pattern — a Claude-powered command center I built to manage client WordPress sites. Site registry, real-time content views, natural-language commands. The same approach goes into your tools.

No demos-as-products. No wrapped chat UIs pretending to be agents. The tools I build get used on Monday morning, without a manual. And if a discovery call lands on “AI isn’t the right spend here” — I’ll say so.

Engagement details

Stack

Claude API · Anthropic SDK · Python / TS

Typical scope

1 internal tool, 1 agent workflow, or a discovery engagement

Timeline

2–6 weeks for a build · 1 week for a discovery

Team

One practitioner (me) + 1–2 stakeholders

Hosting

Your infra, or mine

Pricing

Scoped per project. Starts with a discovery call.

§ 02Process
Discovery → Handoff

Six stages. About 4 weeks for most tools.

01

Discovery

A week watching the actual workflow. Not the workflow you’d describe on a whiteboard — the one with the spreadsheet, the reply-all thread, and the PDF no one opens.

02

Problem pick

We choose one task to replace. One. The boring one that costs the most hours and produces the least judgement. This is also where a discovery engagement can stop, if AI isn’t the right spend.

03

Prototype

A working tool in a week. Ugly, unoptimised, but handling real inputs. We learn more from one bad version shipped than ten perfect mockups.

04

Evaluation

Side-by-side with the human. Eval set built from your actual work. No vanity benchmarks.

05

Hardening

Guardrails, fallbacks, prompt caching, rate limits. The unglamorous bits that make the tool trustworthy when you’re not looking.

06

Handoff

Documented, version-controlled, with a 30-day punch-list window. Your team runs it. I’m on retainer if you want me to watch for drift.

§ 03Deliverables

Everything included.

Workflow audit + shortlist of AI-fit candidates
Working prototype in week one
Eval set built from your real data
Production tool, containerised or cloud-deployed
Prompt caching + retry logic
Guardrails: input validation, output checks, human review
NordDashboard integration (if applicable)
Cost-per-task benchmark
Runbook + handoff doc
30-day punch-list window

Next step

Let’s see if AI earns its keep here.

30-minute intro call. Honest read on whether a tool is the right spend for this workflow — or whether a cheaper fix would do.

Services