The Secret to AI Automation: Why You Should Automate the Edges First
When it comes to implementing AI automation and building AI agents, many teams make a critical mistake: they try to automate the complex core of their work right away. There’s a much more effective strategy that can save months of effort and deliver value faster. The secret is simple: automate the edges first.
[00:04.577]
[A man in a beanie and glasses explains the concept of automating the edges of a workflow.]
“Automate the edges first.”
Most teams burn months trying to automate the core of their work—the very thing that humans are already quite good at. The real leverage, however, often comes from focusing on the peripheral tasks, or the “edges” of a workflow. These are the supporting activities that, while crucial, are often tedious and time-consuming.
[00:17.333]
[A man in a beanie and glasses explains the concept of automating the edges of a workflow.]
The edges include tasks like data preparation, quality assurance (QA), synthesis of information, and the handoffs between different stages or teams. While these might seem like minor parts of the overall process, they are ripe for optimization. AI can quietly compress the cycle times for these tasks by a staggering 70, 80, or even 90%. Despite this massive potential for efficiency gains, most people don’t start here.
[00:26.551]
[A man in a beanie and glasses explains the concept of automating the edges of a workflow.]
It’s important to clarify that this approach is not about picking an unimportant problem to solve. You should absolutely choose an important and valuable problem space for your AI initiatives. The key is that once you’ve selected that valuable area, you should then focus your automation efforts on the edges of the work within it. There are tons of leverage to be found around a valuable problem by tackling the peripheral tasks that support the main workflow.
[00:50.117]
[A man in a beanie and glasses explains the concept of automating the edges of a workflow.]
The “automate everything” vision is tempting, but most core workflows are riddled with complexity. They contain ambiguity, require handling numerous exceptions, and rely heavily on unwritten tribal knowledge. Teams consistently underestimate this hidden state and tend to overestimate the reliability of AI models, especially if they haven’t built an AI agent before. This mismatch leads to stalled projects, bloated scope, frustrated leadership, and disillusioned engineers. Trying to automate the core first is like trying to build a self-driving car before you’ve even invented cruise control.
[01:30.315]
[A man in a beanie and glasses explains the concept of automating the edges of a workflow.]
If this is your first AI agent project, here’s a challenge: pick a valuable workflow you want to automate, figure out the edges, and just test one. See if there’s an opportunity for a quick win that provides a lot of “bang for your buck.” You might be surprised by how much time and effort you can save with a relatively simple automation.
[01:45.897]
[A man in a beanie and glasses explains the concept of automating the edges of a workflow.]
Consider data preparation. How do you currently collect context, clean data inputs, or normalize formats before the core work even begins? Is it a manual process? That’s an edge. Or look at QA. How do you check for doneness, completeness, quality, and consistency? An LLM can be easily trained to perform these checks, freeing up human experts to focus on more subjective assessments.
[02:09.529]
[A man in a beanie and glasses explains the concept of automating the edges of a workflow.]
Synthesis is another excellent example. Instead of automating an entire workflow, what if you just used AI to summarize the discussion thread in a Jira ticket and update the description? This saves everyone time and improves clarity. Similarly, the packaging of finished work—converting outputs into briefs, reports, or presentations—is a super valuable task that often takes a lot of human time but is not overly complex for modern LLMs.
[03:11.455]
[A man in a beanie and glasses explains the concept of automating the edges of a workflow.]
Finally, coordination is an edge that holds a ton of value, especially in environments that rely on tribal knowledge. This often involves someone manually pulling information from one place, talking to someone else, and then putting it in another system. Automating just this piece of the puzzle can streamline the entire process significantly.
[03:32.843]
[A man in a beanie and glasses explains the concept of automating the edges of a workflow.]
So, why do the edges work so well? They are typically high-friction but low-judgment tasks. Because all the necessary inputs are already available, they are perfect for today’s imperfect LLMs. You should assume your first AI agent won’t be perfect, and it needs to deliver value anyway. With edge automation, errors are often easily recoverable and cheap to fix because the humans are still managing the core, high-judgment parts of the workflow.
[04:24.316]
[A man in a beanie and glasses explains the concept of automating the edges of a workflow.]
This strategy doesn’t mean abandoning the goal of automating the core. In fact, it creates a clean path toward it. By automating QA, handoffs, and data preparation, your team gains the deep knowledge and experience needed to eventually tackle the heart of the workflow. You position yourself to snowball these gains across the entire organization.
[05:01.378]
[A man in a beanie and glasses explains the concept of automating the edges of a workflow.]
“You are not just doing a technical project. You are doing an upskilling project not just for the engineers building the agents, but for the humans involved.”
This process is fundamentally an exercise in building trust. By automating around the core craft, you show the human experts that their skills are still valuable. The goal is to help them apply their knowledge more efficiently, not to replace them. This builds the confidence needed for them to share the “secrets of the art”—the nuanced, tribal knowledge you’ll need to successfully automate more complex tasks down the road.
[06:40.485]
[A man in a beanie and glasses explains the concept of automating the edges of a workflow.]
The real leverage in AI hides outside the core, in tasks like intake, data polling, QA checklists, and packaging. When you start there, reliability goes up, risk goes down, and you earn the trust of everyone involved. This is how you create a virtuous cycle that leads to teams winning fast. To get started, pick a valuable workflow, map its edges, choose the simplest one, and build a solution. The workflow itself will reveal the right path forward, balancing human expertise with the power of automation.