As AI agents move from novelty to necessity in today’s enterprises, a critical question is emerging: should we fix our messy processes before deploying agents, or can we rely on the agents themselves to make sense of the chaos? Some leaders argue that process improvement is obsolete and that agents can simply adapt and execute within our existing inefficiencies. Others insist that without clear, optimized processes, automation will only amplify our problems. A case can be made for either, but without proper planning and sponsorship, it’s no wonder MIT found that 95% of all Generative AI projects are likely to fail or yield no value to organizations who are even able to get the projects to full production.
Before diving deeper, let’s acknowledge that process inefficiency is rarely a result of neglect, as it’s usually the product of evolution. Processes that once worked perfectly tend to drift over time as organizations grow, merge, and adapt to new technologies and regulations. What starts as a streamlined workflow can easily become cluttered with exceptions, manual workarounds, and outdated steps that no one remembers questioning. Understanding why processes become inefficient is the first step in deciding whether agents should inherit them as-is or whether we should clean things up first. If you aren’t familiar with the concept of Agentic AI, start with this primer and come back to see why this is becoming a prevalent topic.
Before we go too far, let’s be honest: processes don’t usually become inefficient because someone’s asleep at the wheel or lack of caring. They fall apart because businesses change, people move on, and priorities shift. What worked like a charm five years ago might barely hold together today. As we grow, merge, and bolt on new tech, those once-smooth workflows start collecting duct tape and sticky notes. It’s nobody’s fault; it’s just what happens when reality collides with good intentions. Still, if we’re going to throw agents at our problems, it’s worth understanding why things got messy in the first place.
Why do processes become inefficient to even consider needing improvement? Well, there are a whole number of reasons to consider. Bear with me, as these examples are probably common sense for what you do day to day, but maybe not?
Why Processes Become Inefficient
Let’s start with the obvious one: mergers and acquisitions.
- We live in an M&A world, whether we like it or not. That means trying to combine two or more processes that each company swears is “the right way.” They never match, so we end up debating whose version wins, if anyone even takes the time to analyze it. More often, the acquiring company just declares, “We’ll do it our way.” It’s quicker, but not always smarter. Sometimes it really is easier to just rip the Band-Aid off and move on.
Next is growth.
- Growth is what everyone wants. It looks great on a dashboard and sounds good in an earnings call. But it also creates complexity. The simple little process that worked when ten people sat in one office does not work the same when hundreds are spread across time zones. Over time, we add new steps, checks, and exceptions to deal with changes in products, people, regulations, and customers. Nobody has the time to pause and clean it all up. The layers build until the once-simple process becomes a bureaucratic obstacle course.
Some organizations take pride in constant iteration, always refining and improving their processes. That sounds great in theory, but perfectionism has a price. Every tweak takes time, people, and more change management meetings than anyone cares to attend. I like efficiency and hate waste as much as anyone, but there’s a point where fixing something that technically works can cost more than it saves. At some point, you have to ask whether the juice is worth the squeeze.
Finally, we have edge cases and workarounds.
- Someone adds a manual step “just for now,” and a year later it’s part of the official process. Then turnover hits, leadership shifts priorities, and nobody remembers how it got there. Before long, that one-time exception becomes the rule. That’s when you start hearing the classic line: “We’ve always done it this way.” And that’s the moment you know the process has stopped evolving.
I could go on about new regulations, policy creep, or random technology being tossed in without a plan, but the truth is that inefficiency isn’t a sign of failure. It’s what happens when success, survival, and human nature all pile up in the same place.
So, back to my original question: are we applying agents to deal with the process as it is, or should we fix the process first to make their deployment more effective? Either way, we need to understand what the process actually looks like before agents can do anything useful. Let’s walk through the options, and you can decide what makes the most sense.
The case for analysis:
In reality, we can fix the process, skip the fix and implement anyway, or find a blend of the two depending on the situation. I’ll make the case for each approach, but I’ll try not to go too deep into the details, so you don’t fall asleep before the end.
If you have read anything I’ve written, you’ve probably seen that I put a big focus on efficiency and value. With maximum efficiency comes the requirement to strip out waste. This waste could be that of time, money, manpower, or along the lines of my concern, steps within a process. Sometimes, cleaning up your processes first just makes sense. If your workflows are outdated, overcomplicated, or held together by workarounds, automation will only make those flaws more visible. Agents thrive on structure. They need clarity and rules to follow. When you put them into chaos, they cannot create order; they just scale the confusion faster. In those cases, fixing the process first saves a lot of pain later. Think of it as cleaning the kitchen before buying a new appliance.
On the other hand, agents can be incredibly helpful in finding where the problems are hiding. They see patterns humans miss and expose inefficiencies you did not even know existed. If your data is solid and your systems are reliable, letting agents run inside the existing process can surface insights that help you redesign it more intelligently. Sometimes the smartest move is to use the agents as a diagnostic tool before you start tearing things apart.
What usually fails is the “throw agents at it and hope for the best” approach. It is tempting because it feels faster, and leadership loves speed. But automation without process understanding is like paving a dirt road full of potholes. It looks better for a while, but the bumps are still there. Once the bots are running, those bumps turn into real obstacles.
The key is balance. A quick assessment of process health before implementation does not need to be a six-month consulting project. Just ask a few honest questions. Is the process documented? Are there clear rules and ownership? Does it actually work the way people think it does? If the answers are mostly no, you probably have some cleanup to do before letting the agents loose.
If the answers are mostly yes, then go ahead and bring in the agents. Just make sure you treat their output as a feedback loop, not a magic fix. Let them show you where your process holds up and where it falls apart. Then adjust.
Agents are powerful, but they are not miracle workers. They reflect the environment they are dropped into. If your process is sound, they will make it better. If it is broken, they will make the pain more visible, faster, and louder.
So, if you are deciding whether to fix or implement first, here is how I would break it down in plain terms:
- Fix the process first when the workflow is already a mess. If it is inconsistent, undocumented, or nobody can explain how it really works, automation will only magnify the chaos. Get it stable before you start scaling it.
- Implement agents first when you already have structure but want to identify weak spots. Agents can surface inefficiencies and patterns that would take months for a team to uncover manually.
- Blend the two when the process is mostly sound but could use targeted improvements. You can use agents to help refine, monitor, and even automate the fixes as you go.
These are, of course just some of the thoughts and arguments behind considering cleaning up these unwieldy processes and each organization will have their own. Of course, this requires time and money to evaluate, plan, and fix these existing processes; two resources many of us don’t have enough of. So, maybe it is better to just start applying agents?
In most cases, the right answer depends on your use case, your tolerance for chaos, and your team’s ability to adapt. Every organization operates differently so there is no universal playbook here. The best outcomes come from good judgment, solid data, and a genuine curiosity about what is really happening under the hood.
The case for implementation:
I love technology. I’ve worked in technology for the better part of the past thirty years. Over that time, technology has globalized the economy, revolutionized communication and put information at our fingertips. If you grew up in the 1980’s you would never even fathom the innovation, unless you grew up in one of the tech epicenters such as Silicon Valley or Cambridge. I’m also pragmatic, and don’t believe in technology for technology’s sake. Yes, the newest and coolest tech opens the door for a crystal ball to an amazing future where we humans have robots doing all the work for us and we’re happily on a permanent vacation for some. At the same time, there is such a rush to production, and deployment to satisfy the hungriest of technophiles and reasons to buy the latest gadget, we oftentimes sacrifice quality for advancement. This is a big tangent, so I’ll stop here and discuss it in another article, but I hope you’re starting to see the reason for my concern. Now, leaders see this shiny object that will cut costs, reduce risks and increase revenue. All with the promise of making employees more productive, quickening the order to invoice process, and of course, potentially cut headcount in areas where possible. I don’t love that last part to be honest, but that’s unfortunately the world we’re living in today. Anyway, here are some of the arguments that could be used to take the leap and start deploying.
Increasing speed to value, matters.
- What could really go wrong? We’re already using inefficient processes and it’s taking humans longer than the promise of agents. In rapidly changing markets, waiting to “perfect” processes can delay innovation. Agents can deliver immediate value by automating repetitive tasks, even if the process isn’t perfect. This creates momentum and tangible ROI early.
By implementing agents, we may be able to highlight process weaknesses.
- Automating an imperfect process often reveals inefficiencies, redundancies, or unclear steps that weren’t obvious before. In this view, agents become diagnostic tools, ultimately helping organizations discover what actually needs to change.
Some processes are too complex for manual fixes.
- In many cases, processes are broken precisely because they’re too complex for humans to execute consistently. Intelligent agents can handle complexity more reliably, effectively “skipping” the need for a human-centric redesign.
Iterative improvement Is more cost effective and realistic.
- Instead of aiming for a perfect process first (which can become a never-ending project), organizations can deploy agents incrementally and refine workflows over time based on real-world results. This agile approach aligns with how many successful digital transformations happen.
Agents can uncover and enable new process models.
- Sometimes the introduction of agents fundamentally changes how a process should work thusly enabling workflows that weren’t possible before. In those cases, trying to fix an old process first may be wasted effort if the automation itself transforms the operating model.
Adding the Third Option: The Blended Approach
And then, of course, there’s the blended approach. This is the one that most of us end up using whether we plan to or not. This option accepts the reality that very few processes are perfect and very few agent implementations go exactly as expected. It’s the middle ground where improvement and automation happen side by side.
In a blended model, you stabilize the process just enough to make it functional and measurable. Then you introduce agents to handle routine or data-heavy parts of the workflow while you continue to refine the rest. The agents help surface weak spots, and the humans decide how to address them. Over time, both sides make the other better.
This approach works especially well when there’s pressure to show progress quickly, but you don’t want to waste months analyzing every detail before doing anything. I’ve always been a fan of the “start small, get wins, get budget, and go bigger” approach. It’s also a good option when leadership is skeptical about investing heavily in process work without visible results. The blended path gives you early wins from automation while keeping room for longer-term process improvement.
Of course, it’s not perfect. If you do not set clear boundaries, you risk letting “temporary” workarounds become permanent. That’s where discipline comes in. Treat the blend as a cycle, not a compromise. Implement, observe, refine, repeat. The goal is to evolve both your process and your automation in parallel, not to patch one while the other struggles.
Conclusion: Preparing for the Blend
In the real world, the right answer is almost always somewhere in the middle. Fix what’s clearly broken, automate what’s stable, and use the agents to help you see what you missed. A blended approach gives you the best of both worlds: human judgment guiding structure and automation amplifying insight.
If you are serious about getting this right, start small and stay curious. Map the process as it actually happens, not as you think it happens. Identify where time, effort, or accuracy break down. Then look for low-risk, high-visibility and high-value places where agents can make a measurable difference. And, actually measure the difference, because if you can’t measure it, it’s hard to justify.
Ask yourself three questions before deploying:
- What problem am I trying to solve, and does it need to be solved by automation?
- If I fix the process first, what value do I create that automation can extend?
- If I deploy agents now, what insights can I expect them to reveal that I cannot see on my own?
This type of thinking will save you from the extremes of over-planning or over-automating. It forces you to stay grounded in results rather than novelty. In the end, agents are not here to replace good process design; they are here to make it smarter, faster, and more sustainable. The organizations that figure out that balance will be the ones that see results, not just headlines.
