Your Next Leadership Challenge (And How to Ace It)

Eventually, we’ll all need to become centaurs: half human, half AI. And I bring you the numbers to prove it.

Upwork just ran 322 real freelance jobs through frontier AI models (Claude, Gemini, GPT-5), then had expert humans give structured feedback and let the AI try again. These weren't toy problems but the kind of work your marketing team does daily: campaign copy, content writing, data analysis, lead generation, ad creative, research summaries, and client decks.

The results aren't just interesting. They're a blueprint for what your team actually needs to learn.

On the first attempt, AI-only completion rates hover around 20-40% depending on the model. Add one round of human feedback, and completion jumps by 11-14 percentage points absolute—that's a 29-71% relative improvement. Nearly one in five failed jobs gets "rescued" after a single feedback loop. The leverage isn't in better prompts. It's in a structured evaluation.

The Upwork researchers had expert freelancers turn vague job posts into explicit “rubrics”—think of it as a checklist of what “done right” actually means. They created 5-20 acceptance criteria per job, marked as critical, important, optional, or pitfall. Things like: “All ad variants use approved brand voice guidelines“ or “Each headline mentions at least one concrete benefit.“ Not “high quality work.“ Not “professional output.“ Verifiable, binary checks that take 1-2 minutes each. Then humans scored AI output against those criteria and gave specific, criterion-based feedback. That's the entire intervention.

But here's your leadership problem: this structured feedback skill doesn't exist in the wild yet. Unlike with prompting, your team members can't take a course on “how to write acceptance rubrics for AI output in your specific domain.“ They can't read a blog post that teaches them the difference between useful feedback (“Criterion 3 required UTM parameters; 2 variants are missing them“) and useless feedback (“I don't like this“).

The paper makes clear where AI reliably struggles: deliverable format requirements, spreadsheet structure, financial report templates, precise language constraints, and tracking conventions. In marketing terms, that's channel-specific rules, UTM structures, brand guidelines, word counts, and compliance checks. These are exactly the things your humans should be watching for—and exactly what they need to learn to articulate crisply enough that AI can use the feedback.

The research builds a simple expected-value model showing three zones:

If you don't explicitly define these buckets for your team, you get chaos: ”we trust AI here but not there” based on vibes, not risk. The UpBench data gives you permission to make these categories a formal policy.

The implicit promise in this research: your job is shifting from “doing all the work“ to “designing the rubric, checking the output, and giving feedback that makes AI useful at scale.“

The teams that figure this out in the next 12 months will pull ahead. The ones that keep treating AI as a side toy won't.

— Torsten & Peter

As with any digital ad campaign, the important thing is to reach streaming audiences who will convert. Roku’s self-service Ads Manager stands ready with powerful segmentation and targeting — plus creative upscaling tools that transform existing assets into CTV-ready video ads. Bonus: we’re gifting you $5K in ad credits when you spend your first $5K on Roku Ads Manager. Just sign up and use code GET5K. Terms apply.

Use code GET5K now

We wouldn’t be marketers if there weren’t a Black Friday deal for you.

From today until EOD November 28 (Black Friday), we have a once-in-a-year offer.

The 50% off deal is available only to current subscribers and is live from today.

Go to this checkout page. The discount is already applied to the pricing.

Scroll down to the bottom of our emails (any).

If you are a free subscriber, you’ll see a paywall.

If you are a paid subscriber, you’ll see your access links.

We create a lot of workshops for our Learning Hub members. Deep-dive sessions on AI strategy, prompting frameworks, implementation playbooks—the kind of content that used to take us days to build into a polished deck.

Now? We use Gamma's AI agent to go from concept to presentation in under an hour.

Here's how it works: 

We tell Gamma what we're teaching and who we're teaching it to. The agent generates a full 30-50 slide deck with structure, frameworks, and visuals—ready to refine, not rebuild.

We're not talking about generic templates or slides that look like everyone else's. Gamma handles the layout, hierarchy, and design direction while we focus on the insights. The output is polished enough to present immediately, but flexible enough to customize.

The time we used to spend fighting with slide alignment? Now spent making sure the content actually moves careers forward.

If you're still building decks from scratch, you're spending time on production that should go to strategy.