The Human Bottleneck: How to Build and Automate with AI Agents

This session draws on practical lessons from building award-winning AI systems in days and weeks rather than months, and from running AI agents that work unattended in production.

Most people still use AI one prompt at a time: ask, wait, read the answer, ask again. It's useful, but the person at the keyboard becomes the bottleneck. Agentic tools such as Claude Code and Codex run on the same underlying models but work in a completely different way. Give them a clear plan and they can research, build, test and improve their own work for hours without stopping to ask.

This session sets out a practical method for building almost anything with AI, shows how to turn repeatable work into automated workflows that run on their own, and tackles the question every data professional should be asking: how do you know the output is any good?

Learning Outcomes

By attending this session, participants will discover:

From Chat Assistant to AI Agent

  • Harness engineering: the context, tools and guardrails around a model that decide how far it can go on its own, and how to make that setup improve itself after every run
  • How orchestration breaks a big task into pieces for specialist agents to research, build and review
  • Why the same models give completely different results in an agentic tool than in a chat window

A Four-Step Method to Build Anything

  • Research, Plan, Build, Improve: a method that works at any level of complexity, even in a domain you know little about
  • Why the plan is now where most of the work happens, and what a good one looks like
  • The snowball effect: how every output becomes context that makes the next build better

Automating the Work You Repeat

  • How to take yourself out of the loop: turning a plan into tasks that agents work through unattended
  • What every task needs: the goal, steps to verify the work, where to work and what happens next
  • Turning anything repeatable into a reusable workflow, from software and research reports to video editing

Trusting the Output

  • Is it just slop? Telling good AI output from bad
  • Why AI checking its own work isn't enough, and how to pair it with deterministic checks that give the same result every time
  • Building quality gates and safeguards so you can trust work you didn't watch being done

Target Audience

Data analysts, data scientists, data engineers, BI developers and analytics leaders who already use AI tools such as ChatGPT or Copilot and want to go further, from building their own tools to automating recurring work. No software engineering background is needed.

This masterclass is part of our ‘Member Masterclass’ series delivered by our member organisations on a monthly basis. If you or your organisation would like to run your own masterclass for the benefit of our community, please get in touch.

Kevin Collins

Kevin Collins

CEO & Founder, Echofold