"What should we do when AI does the work?" Mobiinside shared a Meta engineer's answer — twelve tips distilled from less than a year of AI-first work that, by his account, completely rebuilt his personal workflow. The framing is developer-flavored, but the rules apply to anyone whose job is being reshaped by AI, marketers included.
The first cluster is blunt. Being good with AI is no longer a differentiator — it's assumed (tip 1), and the gap is already widening: inside Meta, the best users generate 10–20x the impact while others' productivity barely moves. Becoming "agentic" — understanding prompts, context, token efficiency, and how to encode your company's context for an agent — is where the leverage lives (tip 2). So is using AI to learn (tip 3): the engineer routinely asks agents to visualize unfamiliar systems as infographics, interactive pages, or flow diagrams, which both speeds comprehension and exposes what he missed.
The second cluster is the heart of the piece. Knowing where AI fails is learned by doing, not by reading (tip 4). One habit stands out: when a new model ships, he deletes his prompts and starts from scratch — accumulated prompt libraries matter less than quickly re-learning a new model's failure patterns. And taste is the moat (tip 5), with a definition worth keeping: "Taste is understanding the customer's mind." The details that decide quality rarely live in documentation, which is why — counterintuitively — the AI era argues for spending more time with customers, not less.
The execution rules are concrete. The internal tools that succeed at Meta all shipped before they were finished (tip 6) — with the crucial distinction that "70% polish is not the same as 70% accuracy." Misaligned buttons can ship; wrong results can't. Role boundaries are dissolving (tip 7): PMs code, designers ship, engineers do PM work — through agents. And when you need another team's support, first ask whether an agent can do it instead of waiting in their queue (tip 8).
The rest is direction: aim for the top 1% of your org and have AI propose the ten highest-impact problems to solve, then pressure-test the list with your lead (tip 9); don't confuse using AI with solving problems (tip 10); find the best AI users around you and absorb their workflows (tip 11); and protect your health, because the ceiling on how well you run agents is your own condition (tip 12).
Swap "code" for "campaign operations" and every rule still holds. Delegating creative production and reporting to AI is already the baseline; the gap comes from how well you encode your brand's context — tone, red lines, why past work performed — into the agent. That's precisely the conclusion of What Socar and Samjjeomsam Learned the Hard Way — the AI Gap Is Accumulation, Not Tools. And if your team's AI got faster but outcomes didn't, the bottleneck is usually human review-and-decision points, not the tools — the exact pattern mapped in AI Got Faster — So Why Are We Still Busy? The Four Points Where AI Waits on Humans. The engineer's closing line settles the sequencing question: "The gap is smallest today." Skip the grand adoption plan; solve one of your own problems with AI today.
Two axes carry the piece: mastering AI is now table stakes rather than a differentiator, and the durable edge is what AI can't provide — taste (understanding the customer's mind) and a bodily sense of where AI fails.
Yes — the original explicitly frames them for all roles. Prompt and context design, shipping at 70% polish, ignoring role boundaries, and problem-first thinking translate directly to campaign operations and content work.
No. The engineer distinguishes 70% polish from 70% accuracy: cosmetic refinement can wait, correctness can't. The point is shipping early enough to let real feedback finish the work.
The piece argues the opposite: the agent era has just begun and 'the gap is smallest today.' Starting with one real problem of your own is enough.
To apply what you just read to your own site, start with a free audit of where things are now.
A strategist replies within 24 hours on business days.
Choi So-young has sent a daily AI prompt newsletter since 2023. Her view on the shelf life of AI skills, the taste to judge output, making space to think and why love matters.
As startups scale, early members often stall or leave. Three rules for career growth in a scaling organization: cut operational work, learn wider, act on strategy early.
Feedback like ‘make it bold but simple’ isn’t nonsense; priorities just aren’t set. Four questions that turn vague design feedback into clear requirements.