Waite’s working plan
AI transformation in 30 days One workflow, audited, built, proven, and running in production. Owned by your team. In one month.
SCROLL ↓
Every business has access to intelligence. The same models, available to anyone with a credit card, including your competitors.
If everyone has it, the tools cannot be the advantage. If everyone starts level, what separates you?
Into how you put it to work. Which workflows carry it. Where it decides alone, and where a person stays in charge. That is transformation, and it is a build job, not a slide deck.
Intelligence doesn’t put itself to work. Someone has to build it in.
We do it five ways. Build
Products and agents, end to end.
Embed
We drop into your team for a sprint and ship the hard part.
Architect
We set the AI architecture, build the core, hand it over.
Always-on
An agent army that keeps working after we’re gone.
Teach
We leave your team able to build and run the agents themselves.
Matched to where your business is, not to a package we prefer to sell.
First, see the work as it really happens. The documented process is rarely the real process.
01 A request lands the trigger + →
02 Re-keyed into a spreadsheet by hand + →
03 Checked in the internal system tribal knowledge + →
04 Approved in a chat thread waits on a reply + →
05 Entered into the system of record 40 required fields +
the audit traces every step
Click any step; the “simple” version hides the real work.
Then decide how it should run with intelligence built in. 1 Intake, validation, routing DETERMINISTIC SOFTWARE
2 Gathers, decides, drafts THE AGENT
3 One clear decision A HUMAN APPROVES
4 Record updated, no re-keying DETERMINISTIC SOFTWARE
✓ Evidence log: every step recorded; exceptions escalate to a person ALWAYS ON
We map the current workflow, design the AI-native workflow, then connect the two with a system .
WATCH IT WORK · ONE LIVE RUN
The end point is a system that runs over what you already have. RUN LOG · quote_request.eml
Audit → Build → Prove → Deploy → Observe → Improve
↺ THE LOOP RUNS AGAIN
Waite owns the path from messy workflow to trusted production system . THE 30-DAY PLAN ↓
The 30-day roadmap
One month. One workflow. A team that owns it. Four weeks, each earning the right to the next, on one real workflow, not a lab demo.
WEEK 1 · DAYS 1–7 · AUDIT
Week 1: See the work as it really happens. By day 7: an operating map of the real workflow, one use case chosen for value, boundaries set.
01
Pick the workflow
high volume, repeatable, expensive: the one your team complains about
02
Interview the doers
how it actually works: tools, chats, workarounds and all
03
Trace the systems
inboxes, spreadsheets, the CRM, the thread where approvals happen
04
Find the exceptions
where it breaks, who rescues it, what the rescue costs
05
Quantify it
hours per week, loaded cost, error rate, delay
06
Draw the map
steps, systems, hand-offs, judgment points, failure modes
07
Checkpoint
one use case chosen for value, boundaries set
WEEK 2 · DAYS 8–14 · BUILD
Week 2: Build the agent, on real work, not a sandbox. By day 14: a working agent doing one real job, side by side with your team, every step on the record.
08
AI-native design
each step: deterministic software, the agent, or a human
09
Set the guardrails
validation, step limits, output filtering, before it acts
10
The agent loop
prompt → decide → act, until done or a limit stops it
11
Connect the tools
the systems you already run; no replacement project
12
The audit trail
every prompt, tool call and error, timestamped
13
Run on real work
live cases, with the process owners watching
14
Checkpoint
a working agent doing one real job, on the record
WEEK 3 · DAYS 15–21 · PROVE
Week 3: Turn “it seems to work” into evidence. By day 21: pass rate, failure categories, and cost per run, measured on your real cases.
15
Structured outputs
a defined schema; validate, retry or escalate
16
Failure paths
missing data, dead systems, timeouts: defined behaviour
17
Golden dataset
20 real cases with hand-labelled ideal outcomes
18
Run the evals
correctness, required steps, escalation behaviour
19
Cost per run
cheaper models, caching, token limits: measured
20
Fix the top failure
change one thing, re-run, compare
21
Checkpoint
pass rate, failure categories, cost per run: evidence
FINAL WEEK · DAYS 22–30 · RUN & TEACH
Final week: Into production, and into your team's hands. By day 30: a system carrying real responsibility, and people who can run, read, and extend it without us.
22
Deploy
sandbox first, inside your own infrastructure
23
Monitor & rollback
know when it’s stuck before anyone else does
24
Humans on risk
approvals stay where decisions are irreversible
25
Autonomy dial
more authority only as the evidence accrues
26
Teach: run it
logs, escalations, prompts: the agent becomes theirs
27
Teach: extend it
your people build the next agent themselves
28
The business case
hours back, errors down, against the day-5 baseline
29
Workflow two
the loop runs again: observe, improve, widen
30
Final checkpoint
a running system and a team that owns it
Transform before you announce it. On day 30 you don’t have a plan for AI transformation. You have evidence of it.
THE PROOF
DAY 7 The operating map
DAY 14 A working agent
DAY 21 The evidence
DAY 30 A team that owns it
WHAT YOU KEEP
✓ The operating map
✓ A working agent with guardrails
✓ The eval suite + golden dataset
✓ Cost per run, measured
✓ A trained team
✓ The business case
START THE AUDIT → OR SEE WHAT THE AUDIT HANDS YOU → 