“AI is changing the nature of work” == “AI will make you worth…less” is the mantra that whispers ominously in the quiet lull of a Tuesday night as you have nothing to look at but yourself. I thought this anxiety would hold us hostage in the 40+ zeitgeist-probing conversations I’ve had this past year. But I’m not seeing as many hunched backs and strained eyes to skill up on Claude Code as I thought. That *is *happening, but more calories are being spent on something that’s newer for the data engineers, leaders, and business analysts I’ve been talking with: status capture.

I initially wrote off these scattered anecdotes until the sparks started to cluster:

The modern data stack is over…met with a collective sigh and shrug (read: halo effect is dimming rapidly)

No one is bragging about drilling for oil (remember that old adage “big data is the new…” ah well you get my point)

Experiments with new job titles (think: how Anthropic and OpenAI made “Member of Technical Staff” the new “it” title)

“We want to glue data and internal systems like Netsuite together”

“Can we automate batch purchase orders in SAP Ariba after codifying the request?”

Top of mind napkin math to invest in AI solutions vs. offshore human resources in the Philippines and India (these are overwhelmingly common defaults)? True story: a finance exec told me they’re not interested in cost reduction because they can just pay a couple hundred bucks per month offshore and said point blank, “What new value can you add that we can’t do right now?”

This unlocked a memory of something strange I couldn’t appreciate until now, tinged with anxiety and excitement. I got a job offer that touches every single bullet you’ve read until now…in 2023.

Analytics Engineer – Data Activation:****Offer Letter Contents [It’s worth the click and reading every word. The kind of data Levels.fyi wishes they had.]

Let’s highlight a couple points here that capture the essence of this role.

Develop and maintain data pipelines and models to support workflows in areas such as accounting, payroll, benefits administration, headcount and financial planning.

Examples:

Automating accruals to facilitate a 1-day close

Fully automate sales commission processing

Account intent scoring

Pricing and billing automation

I thought to myself, “Isn’t this just a data engineer with a random smattering of enterprise systems?”

The answer is yes and it’s called an Integration Engineer at Figma and Riot Games, and there’s a flurry of emotions whipping the air. This is hard, tedious, non-obvious path, work (the anxiety). This is perfect for dreaming and being hands on with new AI tools (the excitement). I wipe away the disbelief from my eyeballs coming face to face with the work always there. The tedious, cross-functional, nobody-wants-to-own-it work of making systems talk to each other. The lambda functions duct-taped together and the 1000 nodes in an Alteryx (or Mulesoft, Matillion, Talend, etc.) workflow. The Airflow DAGs that exist because someone in accounting needed a report automated and nobody else would do it (read: no one knows what they want, but they know they want it automated).

This work has always been in the shadows. What’s new is that people are finally giving it a title and a $200k+ salary. What’s new is that companies are competing for people who can do it well…or at least try.

We don’t have to reach into the distant past for loose analogies to learn from. There’s a version of this today with clear boundaries in their problem space. It’s RevOps, Marketing, and Customer Success: salesforce, marketing campaign software, common room, churn analysis. There’s enough of a collective standard hard-won through gnarly battles with apex code, excel pivot tables, and the occasional slack bot that the people at Clay thought,

Why can’t we automate what we already have an aligned intuition for manually?

They got to a point where throwing money and bodies at a tedious, manual problem was a low ceiling game. There have been attempts with point solutions, but they’re too opinionated. Also, they presume too much of how a team grows alongside its processes (I’m looking at you Xactly). Automating these processes is both art and mechanics, and this role drops the façade that it’s either/or.

These are emotional, quality of life problems. A leader can think on first impression, “Making my team toil less doesn’t add value to the bottom line.” But in this instance, it literally does because time kills all deals. There are enough shared moments of signing the contract at 11:59pm to squeeze in a deal that everyone knows was delayed by manual deal desk ops. There are enough shared moments where churn could have been prevented if there was better signal automation. Shared suffering catalyzes shared alignment.

The GTM story sounds familiar, right? It sounds like our lives before dbt, before redshift, before column level lineage and all the tools we take for granted. The kind of experiences so baked into our cultural substrate that it’s weird to brag about using them. But we all know with primal disdain how much doing data work sucked before these tools were normal…before Analytics Engineer was normal.

We’re seeing the north star shift before our eyes. Build the pipeline, model the data, serve it to the visualization layer, declare victory. This was the canonical success story of the modern data stack era.

It’s also exhausted. Everyone knows it’s exhausted. If you know…you know how much our necks ache hitting the low ceiling of this experience. The dashboards cover 80% of what people want to know, and the remaining 20% (read: the questions that actually matter, the ones that change decisions) those get answered in excel pivot tables, slack threads, and behind closed-door game-theory-laced conversations (think: “Sent from my iPhone” energy). We all know this. And now that everyone’s personal life is full of LLMs answering ad-hoc questions instantly, the expectation has shifted. The tolerance for “wait for the next dashboard refresh” is approaching zero.

The first brick is being laid: metrics layer data analyst Slack chatbots. They’re clunky now, but the direction is clear. Business users want *way more *agency over data. Data Engineers want way fewer ad-hoc requests. Leaders want legible analysis on demand. The incentives converge because time kills all deals progress.

What’s next after chat? It’s something you’re already doing (or at least view as normal): Reverse ETL. Ingest data, transform it, push it back into the operational systems where decisions actually happen. Salesforce, NetSuite, Stripe…not dashboards. Whether it’s Hightouch, Census, or making your own integrations, there’s a shared worldview this is valuable behavior. It’s brewing for its moment to be mainstream behavior.

I thought of a lot of ways in writing this like a tools and persona matrix. But that doesn’t capture the heart of what we’re aiming for. We’re straining for new stories because building a career on being the “XYZ tool guy” doesn’t have the same cachet in the social capital markets now.

Path One: I want to be a Software Engineer, Data.

There’s a mantra for data engineers to “model their behavior after software engineering best practices.” The next step is to cut out the abstraction and go straight to the heart of it: being a software engineer. This is for the cohort of you all coming to a crossroads in your career: to maintain a collection of data pipelines (that feel in essence point solutions) in airflow or build a reputation for building systems. And more precisely, operational systems. The great data engineers I’ve had the honor of working with see the craft’s endgame in its current form. It’s to canonicalize a company’s incentive structure. What charts make us behave in ways to envision promotions as a secondary effect of a graph going up and to the right? What dashboards make leaders nod their head 99% of time and make them whip their necks down a rabbit hole (think: churn risk) of *productive *detective work? You build a language of power, and there’s pride in that craft. However, I know there’s a minority cohort that aren’t satisfied with this. They want to be causal to the graph going up and to the right.

This is where you’ll see AI as a reason to “go back to school” by learning on the job. It’ll sound like this to start: What if I inject data to serve user-facing, in-product analytics? What if I just build a robust chatbot that tangibly feels like more than text to SQL? What if I build a data and integration platform that centralizes the maintenance and ergonomics of these APIs? What if I build a system to dynamically build new features and fix bugs based on usage analytics and logs?

Then comes the how: Do I want to become good at distributed systems (think: the pain you feel babysitting airflow pools multiplied)? Do I want to learn about durable execution? Do I want to leave DAGs? Do I want to take granular observability seriously? Do I want to care about millisecond performance and Big O Notation? Do I want to think in CAP theorem constantly? Do I want to babysit Kubernetes clusters? Do I want to learn FastAPI? When should I consider vector embeddings?

It’s a lot…because it is a lot. For those of you that see net positive excitement against the backdrop of anxiety, you’ll have a new mantra: I solve problems through systems doing whatever it takes. I build products. This isn’t a prediction. This is already happening now at Anduril (I introduced them to SQLMesh!)

Path Two: I want to be a Data + AI Engineer/Context Engineer.

You love building data pipelines. You love pumping clarity into the mouths, eyes, ears, and brains of every person in the company for which way move to past the fog of war. The software engineering track isn’t what you want. You’re all in on evolving what data engineering means: worldviews and incentives. You see all this AI hype, but you know better than most that data modeling is art and logic. You’ve adopted cursor or claude code and see how it makes you faster, but you’re pondering how it’ll make you better.

This is where a large cohort will start to think: Huh, I noticed airflow added a human in the loop task: HITLOperator. Maybe that’s a good excuse to build manual review for business user intervention for department-specific DAGs? Huh, I noticed a lot of business users are downloading to excel to inform how it will do these things: pricing and billing automation, account intent scoring. What if I just automated that away? What if I inject data to serve user-facing, in-product analytics? What if I just build a robust chatbot that tangibly does more than text to SQL? What if I evolve the supply chain of information vs. managing a collection of solutions ranked by department complaints?

Then comes the how: I wonder if anything has reached a “dbt moment” for a canonical framework? LangChain, pydantic-ai, native python SDKs from OpenAI/Anthropic, or a combination of claude code and beads. Should I build custom integrations or use a ReverseETL platform like Hightouch or Census?

Your new mantra:** I build canonical worldviews for the company and customers. I drive what’s worth caring about.**

You’ll notice an overlap in problems they tackle, but there’s a difference in how they solve them. One identity is building products while the other is hardening and expanding a supply chain.

We wait with bated breath to reinvent ourselves, but no one wants to look cringe doing it. In plainer feelings, being a beginner again feels like a death sentence (even when we intellectually know we can only be beginners at this technological inflection point). And with employer to employee power asymmetry still so lopsided, there’s merit in being a faster horse than doing the hard work of becoming a car.

This shared anxiety unites us. I smell the intersection of top-down incentives to be efficient with bottoms-up incentives to inject new prestige in this role. We don’t want to be seen as simply moving data from point A to B anymore. There are enough medium-sized bets where we can say, “It’s worth trying something new, not just incremental.”

We’ll see erratic behavior to take advantage of this prestige vacuum. There will be leaders who will pigeon-hold this to internal-only systems and gate keep customer-facing systems for “real” engineers. There will be departments that try to hog you and turn you into their shadow IT ad hoc request guy. We’ll see the Big 4 and large consulting firms fight like hell against the erosion of billable hours given how efficient you’ll become. We’ll see a proliferation of CLIs and integration frameworks designed for agent interoperability which will then inspire new startups to build platforms for this rising trend. We’ll see people reinvent the data orchestrator flavored for AI workflows: hatchet and dbos. We’ll see existing orchestrators do the same. We’ll see Fivetran push to solidify this new wave of behavior to get returns on their back to back acquisitions (Census, dbt Labs, Tobiko). We’ll be waiting for FAANG and AI companies to canonicalize a new title to model after. We’re waiting for the series of blogs from big tech bragging about what they did so the rest of us can copy/paste.

We’re waiting for permission. But the work isn’t waiting. It’s hiding in offer letters that don’t quite fit existing categories. The latent demand is manifest demand.

We’re just looking for a name to agree on.