Across the Proxify network, we have seen the same tools lead to very different outcomes depending on how they are used. That is why we are focused on enabling engineers to build agentic workflows that hold up in practice, adopt more effective ways of working, and turn AI into real productivity gains and stronger delivery.
The performance gap is changing
For a long time, the question we asked about an engineer was relatively simple: how deep is their technical knowledge?
That question has not gone away, but it is no longer the whole picture.
Two engineers with similar technical depth can now produce very different results, and an increasingly important part of that difference is how they work with AI.
As agentic coding tools become more capable and more widely available, simply having the tools is becoming less of a differentiator. The performance gap is increasingly about how engineers work with agents, not whether they use them.
Access is not the differentiator
Over the past months, we have run 281 expert sessions with engineers in the Proxify network, looking closely at how they use coding agents in their daily work.
What stood out was how different the results were, even when engineers were using the same models, the same editors, and the same documentation, with no clear connection to seniority.
In one session, an engineer who had already been using AI coding tools for several months watched an expert demonstrate a more advanced agentic setup and said: “I thought I was getting a lot out of AI. Apparently I’m barely scratching the surface.”
That reaction captured a pattern we saw repeatedly.
Engineers were already using AI, sometimes extensively, but many had not yet built the habits and systems that make it consistently useful over longer stretches of work. Seeing a stronger setup in action often revealed how much more they could get from the same tools.
From using AI to working agentically
Good agentic engineering is not simply about prompting more often or generating more code. It is about setting agents up to work well over time.
That means giving them the right context, clear project-level instructions, well-structured tasks, and a reliable way to check the output.
Our hands-on sessions showed that these practices were far from consistent. Among the engineers whose setups we could clearly assess, around half had no reliable way to carry context from one session to the next. Every new conversation effectively started from zero.
That was the single biggest gap we measured.
Most engineers knew these capabilities existed. The challenge was turning that knowledge into a repeatable way of working, rather than getting good results from an agent one session at a time.
Why we built Agentic Fast Track
The same questions kept coming back, in different forms, across those sessions: do you have a template I can copy? A repository I can study? Somewhere practical to start?
Agentic Fast Track became the answer we did not have.
It is Proxify’s open playbook for agentic engineering: the setups, daily practices, and lessons shaped by our team’s performance enablement work, the tools we have built, and what we have seen across the network.
The aim is to make those practices openly available rather than keeping them behind internal knowledge or individual expertise.
But publishing the playbook taught us something important. Engineers could read it, agree with it, and still go back to working exactly as before.
Reading about a craft does not build it.
The ADKAR model helps us understand why. Awareness, desire, and knowledge were often already there. What was missing was ability, which comes from doing the work in your own environment, and reinforcement, which comes from doing it again.
Value doesn't come from introducing the technology. It comes from turning new capabilities into repeatable ways of working that improve performance.
Building the practical layer
That is why we added a practical layer around the playbook.
It combines expert feedback with hands-on practice in the engineer’s own environment. The aim is to identify the areas that would make the biggest difference, then turn those areas into concrete changes through the dojo.
The focus is practical: carrying context between sessions, writing instructions the agent can actually use, and structuring work so the agent can operate independently for longer.
The important part is that the setup has to work beyond the session itself. Use a fresh agent that has never seen the original conversation to check it.
“If your setup only works because you just explained everything in chat, it fails.”
That is the difference between “it worked once for me” and “it works.”
These are exactly the habits that can change quickly once engineers see a stronger way of working in practice. Over the summer, having a project instruction file became the norm rather than the exception among the engineers in the network.
The file itself is not the point. The shift in how they work is.
The point was always high performance
Ultimately, none of this is about increasing AI usage for its own sake.
A token leaderboard will not make an engineer better. Neither will giving them access to another AI tool. Tokens, prompts, and time spent inside an agent tell you how much the tool is used, not whether the work improved.
The real question is whether engineers can understand an unfamiliar codebase faster, move from an idea to implementation with fewer handoffs, give agents enough context to operate independently for longer, and check the result without introducing additional risk.
Those are the outcomes that matter.
High performance rarely comes from one tool or one individual behavior. It comes from making effective practices repeatable.
“Strong agentic engineering skills can have an impact beyond the individual engineer. They can raise the way the whole team works.”
Better instructions, clearer context, stronger checks, and reusable workflows do not stay with one engineer. They spread across the team and improve how work gets done.
Over time, that is how agentic engineering becomes an organizational capability rather than an individual advantage.
