AI & Agent Readiness

AI & Agent Readiness: AI that helps without holding the keys to production

An AI readiness assessment, or a hands-on build, to bring AI agents into your product and operations. The focus is on where AI actually pays off, and on guardrails that make the wrong action impossible rather than merely discouraged.

Is this right for you?

A good fit if

  • You want to add AI features or agents but are unsure where they pay off
  • Your team is already experimenting with agents that touch real data or systems
  • You sell to customers who will ask how your AI handles their data
  • You work in a regulated area such as healthcare or finance

Not the right fit if

  • You want a chatbot added to the website for marketing reasons alone
  • You need research into training new AI models

Where AI projects usually go wrong

  • Agents are given broad credentials because it was faster to set up.
  • Safety rules live only in a prompt, which the model can ignore.
  • Nobody can say afterwards what an agent did, or why.
  • AI is added where a simple rule or automation would be cheaper and more reliable.
  • Customer data flows to AI providers without a clear policy.

What you get

  • Where AI and automation will actually pay off, ranked by value and risk
  • Least-privilege access and approval gates for agents
  • Audit trails that satisfy SOC 2 and HIPAA reviews
  • Data handling rules for AI providers and models
  • Agentic workflows designed end to end, or built with your team

How it works

  1. Discover

    List the workflows where AI could help and what each one touches: data, systems, money, customers.

  2. Rank

    Score each use case for value, risk and effort, and pick the first ones to build.

  3. Design guardrails

    Define permissions, approval steps, logging and fallbacks before anything reaches production.

  4. Build or guide

    Build the first agent workflows with your team, or hand over a design they can build.

What changes for you

  • AI work focused on the few places it makes a measurable difference
  • Agents that can only do what they are allowed to do
  • A clear answer for customers and auditors about how AI uses their data
  • A repeatable pattern for adding the next AI workflow

Related work and writing

The problems are common ones.

Logistics platform

The problem
BookMyCargo, a B2B logistics platform, had to support high-volume, multi-country operations, distributed vendors and real-time coordination.
What I did
Re-architected it as a cloud-native, event-driven platform on AWS, added AI-assisted exception handling and rule engines, and set CI/CD standards for a distributed team.
The result
High-throughput shipment processing with far fewer manual steps, and predictable releases across borders.
Read the case study

Healthcare system

The problem
A healthcare product needed to handle patient data where failure and data leaks are not an option.
What I did
Designed a HIPAA-compliant architecture with defined, audited data flows and access controls built in from the start.
The result
Compliance built into the architecture from day one, instead of bolted on later.
Read the case study

Agents need permissions, not instructions

A sentence in a system prompt is a suggestion, not a constraint. If your agent can reach it, your agent can break it. Build guardrails that make the wrong action impossible.

Questions about AI & Agent Readiness

What is an AI readiness assessment?

A review of where AI and automation can help your product and operations, what data and systems they would touch, and what has to be in place first: access controls, data policies, logging and the right architecture.

How do you keep AI agents from doing something harmful?

By limiting what they can reach, not by asking them nicely. Agents get the narrowest permissions that work, sensitive actions need human approval, and every action is logged so it can be reviewed.

Can AI agents be used in HIPAA or SOC 2 environments?

Yes, if the architecture is designed for it: controlled access to protected data, agreements with AI providers where required, and audit trails that show who or what did what, and when.

Do we need to hire an AI team first?

Usually not. Most useful AI workflows are built by a good product engineering team with the right design and guardrails. The assessment tells you whether you need specialist skills.

Tell me what is breaking.

I will tell you honestly whether I can fix it.