Welcome to Management

Travis and I have been in the trenches at Limbic Systems, deploying real agentic business systems that are rapidly and demonstrably changing how much our clients can get done. Bluntly, the bet we made on our thesis about how the future of work has changed has had a far greater impact in far less time than we anticipated.

One of the first statements we make to the room during our routine onsite intensive is this: “Congratulations, you’ve all been promoted to management.” Their response is waves of nervous laughter, every time.

From the start, we intentionally chose to focus our energy on helping small-to-medium-sized law firms deploy AI to transform how they get work done. One of the things that compels us to care about the SMB sector is that they’re often overburdened with a greater breadth of jobs that must be done outside of their “billable hour” work. For example, a staff of lawyers practicing trademark law spends a major chunk of their time doing non-legal work to keep the lights on, grow their business, and support their employees.

As I sit with our clients and demonstrate what we believe is today’s blueprint for the future of work, the mood in the room is a heady mix of excitement about the possibilities, unease about the future of their careers, and intimidation by its vast potential, among others.

I imagine it’s akin to how those lucky few business owners felt the first time they used a personal computer and realized it would fundamentally alter how they worked from that point forward. This time, we’re the lucky ones.

A common refrain we hear is that managing AI feels a lot more like managing people than they expected. They find themselves checking in on how work gets done (tasks, timelines, outputs) and deciding what’s good enough to move on.

Helping these firms adopt a completely new way of working has also profoundly helped me. It’s forcing me to continuously rethink my own relationship with technology, software engineering, the computer, and the role I play in getting work done. I find myself wondering just how many ego deaths I can experience in a single quarter.

I grew up in Silicon Valley defining a lot of my identity around being a great software engineer, so I have a strange relationship with the advent of agentic coding models. Clearly, they cannot be ignored, and they are not going away. That means our relationship as producers of knowledge work must indelibly adapt and change as well.

Of course, it’s foolish to blindly accept what comes out of any black box, and I’m not advocating that. However, contemplating what it would take to earn that level of trust has led me to reflect more deeply on the past roles I’ve played as a black box to former leaders and bosses who trusted me implicitly in the very same way.

How did they know I was doing a good job? Did they care as much as I did? Should I have cared as much as I did? What would have happened if I did the wrong thing? How would they know? These questions pulled me away from the craft I fell in love with and toward how I measure the impact and value of outcomes.

I've been here before

This AI-fueled emotional fugue state felt new to me at first. However, I remembered I’d been through it once before: when I chose to shift the focus of my career toward engineering management.

At some point, I had a clear sense of my limitations as an individual contributor. My engineering work was more than fine, but as my ambitions grew, I knew I would ultimately be limited if I didn’t invest in coaching, planning, and growing others around me to do great things, collectively. My time at Google allowed me to grow these skills, and I am forever grateful for the mentors I had and the lessons I was paid to learn there; that imprinting will never leave me.

Plenty of great engineers never make that jump. After all, writing code is easy; managing people is hard. Stepping back from the work feels like losing touch with the thing you love. I get it; I felt it too. But AI is making that choice for all of us, whether we like it or not. I watch it happen in real time.

What I didn’t anticipate was how crucial that investment in management would become for this new way of working. Watching associate lawyers (right out of law school) get thrown into this AI rift is wild and overwhelming, to say the least, and I really feel for them. It’s equally jarring for the seasoned attorney who’s never had the chance to manage anyone in their career. That’s why we open with the promotion, and why the laughter is so nervous.

All of this reminds me of Marshall Goldsmith’s adage: “What got you here won’t get you there.” For these skilled knowledge workers to be effective now, it’s essential that their training, judgment, and expertise are constantly leveraged to validate, steer, and confirm that the outputs of large agentic workflows meet (and ideally exceed) what they could deliver themselves, at a speed they couldn’t compete with.

Yet we expect them to do far less of the hands-on work and still be responsible for the results! It could never be as good as me, right? Welcome to the paradox that your boss experiences daily managing you. It’s nice to have you on the same page.

How your boss managed

So, how did your boss do it? (Assuming they did it well; not all of them did.) I can think of a few things that have worked in my experience and continue to produce success with agents:

Own the system that produces results

Effective managers spend most of their energy on how work gets done instead of doing the work. After all, lawyers don’t go to law school to not practice, right? Pretty much every knowledge worker needs to read High Output Management (and arguably Unreasonable Hospitality, if you know, you know) immediately. If you haven’t, this is a priority. I won’t waste keystrokes reiterating it.

Focusing on the systems of how work gets done forces you to consider every part that produces something of value (and the waste byproducts, too). This requires effort; in fact, it’s a full-time job, and handing our work to agents both makes room for it and demands it.

Being responsible for the system is also about ownership. Your reputation should be on the line, and that should matter to you. Accepting bad results from agentic work is your choice. Do these models produce garbage? Yes, of course they do; so do humans, fairly regularly, for any number of human reasons. Do we accept this from our peers? I don’t.

Agree on tolerances, then tighten them

Tom West’s maxim from Tracy Kidder’s The Soul of a New Machine, “Not everything worth doing is worth doing well,” has been burned into my mind ever since I read it. It couldn’t be more important in the agentic era. Models are notoriously bad at understanding where it’s safe to cut corners; seasoned professionals have that hard-earned wisdom, and they can easily convey those lessons to steer agents.

One technique I deployed at Google with consistent success was getting my engineering teams to agree on a “warranty” for our projects. The exercise calibrated people’s expectations for what good enough is, where we should spend our focus and where we shouldn’t. How many users should this support before we need to throw it away and start over? How many seconds can an operation take before someone gets annoyed?

Tolerances exist on a spectrum, and ideally, we’re falling near the center of our prescribed range. As we get better at this, we earn the right to tighten that range, because we have something we can evaluate: a criterion we can observe and a target we can point to that steers us away from over- or under-engineering a solution.

Define what success and failure look like

As someone who was drawn to engineering because I enjoyed solving real problems for people, I accidentally fell in love with the craft. However, prioritizing the love of my craft felt, in many ways, like a selfish act. Does the end user have a better time using my work because I wrote it beautifully or optimized it in a way I found intellectually satisfying? Probably not.

Much like tolerances, the better we are at describing what success and failure look like for our work, the more mastery we have over the problem space we’re responsible for. It forces us to think more deeply about why we’re even doing something. For agents, this is immensely helpful too; these guardrails help agents “hill climb” for longer and longer sessions because they have a clear target.

Define problems and constraints, then let go of the solution

This one was the hardest for me. As an engineer, I was initially extremely pedantic and overprescriptive about how I like things in codebases I’m responsible for. It took me some time to get over this one when working with agents.

Since shifting more of the responsibility for execution to models, I’ve relaxed my preferences, biasing toward what’s easier for the model to manage. Where I still have constraints, I enforce them mechanically, with linters and formatters that run automatically and continuously to apply the style preferences I believe are important. This is no different from working with a group of engineers; teams adopt coding style conventions, error-handling practices, etc.

Focusing on problems and managing within constraints is one of the de facto responsibilities of a good manager. You don’t have the budget, you don’t have the headcount, you don’t have the time you need. Constraints like these dramatically narrow the set of available solutions, and we rarely control them. Describe them clearly to your team or your agents, and let them find the path.

Where are we headed?

Since we’re buying into the idea that managers aren’t really doing anything (lol), what am I, the former doer and now full-time manager of an army of agents, supposed to do with myself while all my work gets done for me?

I encourage you to use this newfound whitespace to bring surprise and delight into your sphere of responsibility. You now have time to “make it nice” (Will Guidara would be proud). We now have the capacity to scale our ambitions beyond what we could ever have done alone. Nobody should be asking you to accept less from agents, and you can choose to ask more of yourself with this new responsibility.

Nobody ever said it was easier to get exceptional work out of your team than to do it yourself; that challenge hasn’t changed at all. However, if we trick ourselves into thinking we’re faster alone, this moment will pass us by, and the people who understand what this pivot in responsibility requires of them will advance past us.

You can keep fighting the good fight as an individual contributor, but the prize for winning the pie-eating contest is eating more pie. Have you had enough to eat yet?

About the author

Derek Perez is the Founder & Principal Technologist of Limbic Systems, a software consultancy that partners with organizations to design and deploy AI-native software and systems — without losing the human judgment that keeps them grounded. After over a decade of hands-on experience building startups and leading high-performing teams at Google, he partners with founders and operators who need both technical depth and business acumen to navigate what's next. Find him on LinkedIn and Bluesky, or by email: derek@limbic.systems.