Why has AI turbocharged coding but mostly produced workslop everywhere else?

Coding is highly structured, while most knowledge work isn’t.

AI won’t supply that structure, the person using it has to.

In just a few years, engineers have seen their work transformed. Many barely type code anymore, and the most sophisticated manage fleets of agents and can 10-100x their output. At RevelOne, we’re building a search intelligence platform with just a couple of developers, a project that had previously been out of reach for a firm of our size. 

That shift happened because code comes with structure built in.

• Requirements are specific,
• Success is measurable, and
• Code either runs or it doesn’t. 

Engineering had built a process around that structure long before AI: specs, tickets, tests and code review.

Most marketing and GTM work lacks this structure. 

• Researching a market, 
• Producing content and managing data serve clear goals: winning customers, retaining them, and
• Forecasting accurately. 

But the steps in between are fluid, and the success of the end deliverable is hard to verify.

AI accelerates us in two ways: more output faster and specialized superpowers, from writing to image creation to data analysis. With knowledge work, we went from a paintbrush to a 20-nozzle paint cannon. It’s no surprise that things got messy.

The same tool can make a mess or multiply real expertise depending on how a person uses it.

Let’s step through two familiar go-to-market tasks, one qualitative and one analytical, and how each plays out.

‍Example #1: a Product Marketer doing Market & Competitive Research

Pre-AI Diligent Specialist:

• Reads industry reports, company sites and customer reviews, hears from customers directly through win/loss interviews, sales call recordings and time with reps
• Over two weeks, builds a clear view of three segments and five competitors: who to target and where the product wins
• Has to prioritize because they can’t cover every adjacent market or every way to slice the segments
• Writes a concise report with conclusions they can explain and defend

AI-Accelerated Mess:

• Avoids customer calls. Makes confident claims about customer needs based on AI-generated summaries, AI-assembled market sizing mixes numbers from different definitions of the market
• In three days, produces 12 documents on segments and competitors, including ten competitor profiles with the same generic strengths and weaknesses. The writing is awkward “LLM-speak”
• Has no view on which segments or competitors matter, so every document ends in a dozen unranked recommendations.
• Having only skimmed the documents, can’t answer “which segment do we lead with?”

AI-Amplified Expertise:

• Talks to customers and reps firsthand, then uses AI to mine 200 win/loss and sales call recordings for patterns no one could hear by listening alone.
• Before any research runs, writes the outline: the segments, competitors and questions that matter, using AI to test and refine it.
• AI agents research hundreds of sites and data sources. The PMM checks the sources behind key claims, makes sure the market sizing uses one definition of the market, and follows up on contradictions.
• Screens eight segments instead of three, goes back and forth with AI on the promising ones, and narrows to a few insights that change the recommendation.
• Writes the summary and recommendations themselves, and can say which segment to lead with and why.

Example #2: a RevOps lead doing Pipeline Reporting & CRM Cleanup

Pre-AI Diligent Specialist:

• Knows Salesforce, the business definitions and why the odd records exist (the deal booked under a parent company, the stage one team always skips)
• Over a 20-hour block, builds four dashboards on one agreed definition of pipeline and catches 15 deals with data issues.
• Has to prioritize because they can’t clean thousands of records by hand or build every analysis the team asks for.
• Can explain every number in the reports and fix the problems they find.

AI-Accelerated Mess:

• In an afternoon, produces a 17-tab spreadsheet with 10+ cuts of the analysis and pages of methodology nobody reads.
• Builds 15 dashboards, each with a slightly different definition of “active pipeline” and “conversion rate,” some invented by the LLM.
• A chart shows win rate falling 8 points. The analysis had quietly redefined the stages.
• An agent “cleans” thousands of records, but overwrites actual exceptions and mis-associates parent companies to deals, so the CRM looks tidier but is less accurate.
• Can’t explain specific numbers or choices, and nobody knows which output is the source of truth.

AI-Amplified Expertise:

• First writes down the definitions (active pipeline, conversion rate, stages) and the known exceptions, using AI to test and refine them.
• Designs the process with AI, deciding which steps follow fixed rules and which need the AI’s judgment, and builds in QA checks.
• Tests a small batch, reviews it, and asks the AI about ambiguous and low-confidence cases before running across 10,000 records and 20 fields.
• Spot-checks each type of record, including the known edge cases, and has AI draw a diagram of the process so others can follow it.
• When a number moves sharply, like win rate, traces the cause before it goes on a dashboard.
• Writes the summary and recommendations themselves. The result is cleaner records at scale, one set of dashboards the team trusts, and a few findings it can act on.

In both examples, the person brings structure and judgment to the steps that matter. They have to “own the work” however it was produced. When someone can’t explain the terminology or methodology in their own work, I stop the conversation. We’ve seen long analyses on our own team that the author couldn’t walk me through, and candidates whose AI-written case studies listed sourcing segments they couldn’t explain.

In practice, owning the work comes down to picking the spots where you step in with your own judgment. Here are three principles I’ve seen work.

1. Definition before output

Define work specifically before you delegate it. One of the biggest sources of trouble is jumping from a broad topic or high-level questions straight into a prompt or an agent that generates a long deliverable. Once the AI runs ahead of your thinking, you end up on the outside, working out logic and structure instead of setting them. Before anything runs, write down the goal, your assumptions, the key terms and a rough plan of the steps. Everything that follows depends on it.

2. Iterate, don’t “one-shot”

Even with a detailed plan, don’t hand the AI the whole job and take what comes back. That “one-shot” approach lets errors compound out of sight and again pushes you away from comprehension and ownership. Break up the work and pick spots to guide and interrogate. Ask follow-up questions, test assumptions and spot-check as you go. Steer and redirect the way a good manager does with a team member.  

AI is also eager to agree with your ideas and to overdesign a process, so push back. Three questions I like to ask it along the way:

• Have we overcomplicated this, given the goal? 
• What are we not thinking of? 
• What are two ways to make this plan better?

3. Checks that fit the work

As AI raises the speed, volume, and autonomy of the work, the checks have to scale with it. Human spot-checks of logic and data still matter, but they can’t keep up with thousands of records or hundreds of customer sub-segments. The same AI that can make the mess can also check the work at a scale and level of detail that wasn’t possible before. It can cross-check an analysis five ways or trace every claim in a report back to its source. Unlike a person, a new AI session has no pride in or memory of a deliverable it’s reviewing. Designing these checks is becoming a skill alongside the work itself.

There’s no single formula here. Each project needs its own mix of iteration and human involvement. 

All of this is a moving target, because the water line of AI capability keeps rising. In a few years we went from a “super-Google” that answered one prompt at a time to agents that run entire projects on their own. Like employees who mature, agents will take on more autonomy. Like leaders of a growing organization, we’ll manage more of them. What won’t change is each person’s responsibility to set direction, check what matters and defend the results.

We’ve been talking about outputs and quality, but working this way also matters for the person doing the work. Handing off work wholesale is alienating, and your expertise quietly erodes. Engaging with the scale and opportunity these tools open up does the opposite. You keep learning, and you can take on problems more complex than you could have handled before.

About RevelOne

RevelOne is a specialized go-to-market search & advisory partner that drives Growth through People. Growth strategy and talent strategy are completely intertwined, yet often handled by different people. We staff projects with expertise across both to support our clients in sharpening their growth plans and ensuring they have the right full-time and part-time talent to achieve their specific goals.  

Over the past 10+ years, we’ve successfully placed 2,000 people at over 750 clients, including both tech companies and traditional companies looking for modern GTM leaders. Over 50 of these clients are now unicorns.

Our GTM retained search practice focuses on Marketing, Sales, Client Success, and Partnerships/BD permanent hires for all levels, from executives to directors, managers, and team buildouts. We can also source temporary hires – pre-vetted GTM experts – for strategy and execution on interim, part-time, or project-based engagements.

Contact: Have a GTM question, a new hire, or a problem you’d like to solve? Reach out to RevelOne today to discuss: dweiner@revel-one.com