The Warning
On September 17, 2026, Shopify cofounder and CEO Tobi Lutke went on The Knowledge Project podcast with a sharp message for his own employees: the company's biggest AI failure mode isn't underuse. It's careless overuse. He called it "slop grenades": polished-looking AI emails and code handed off without a human ever actually reading them. "We call those 'slop grenades' that people toss at each other," he said. "That's definitely a bad thing." (Fortune, September 17, 2026)
The warning matters for any marketing team building content with AI, because Lutke is naming a productivity leak most dashboards miss. The cost doesn't show up in the sender's tool. It lands in the reviewer's calendar.
What Slop Grenades Actually Are
Lutke is not criticizing AI itself. He is criticizing the workflow: someone drafts an email, report, or chunk of code with AI, skips review, and ships it as finished. The receiver then has to reverse-engineer the intent, fact-check claims, fix broken links, and rewrite sections. That is the leak.
The behavior has formal data behind it. In 2026, BetterUp Labs and Stanford's Social Media Lab surveyed 962 full-time American desk workers and coined the term "workslop": AI output that reads clean but lacks the components that would make it useful (Fortune, September 17, 2026, citing the BetterUp/Stanford study).
| Finding | Number | Source |
|---|---|---|
| Workers who admitted sending workslop to colleagues | 52.7% | BetterUp Labs + Stanford Social Media Lab, 2026 survey of 962 U.S. desk workers |
| Workers who reported receiving workslop | 38% | Same study |
| Monthly time spent reworking received workslop | 3.4 hours per recipient, up from 2.0 hours last year | Same study |
| Monthly cleanup cost per single employee | $186 | Same study |
| Annual cleanup cost, 10,000-person org | Up to $9 million | Same study (Fortune's framing) |
| Recipients who wanted to avoid working with senders | 36% | Same study |
Those numbers describe a real budget line that never shows up in any AI tool's ROI dashboard.
Why The Failure Mode Shifted
A year earlier, in 2025, Lutke told Shopify employees that using AI was "a baseline expectation" and that they should prove they "cannot get what they want done using AI" before asking for more resources. That pushed adoption. The 2026 correction is the predictable second-order effect: when AI production becomes mandatory, output volume grows faster than the quality controls around it.
Duolingo CEO Luis Von Ahn described the same pattern. After declaring Duolingo "AI-first" in 2025, he told Fast Company in May 2026 that scaling AI content surfaced "pure slop" in roughly 20% of cases: "Whenever we scale a lot [of] things with AI, we have to really be careful that slop doesn't get through." (Fortune, September 17, 2026; Fast Company, May 2026)
The shared lesson: AI bonuses only cash out when the workflow includes a human checkpoint.
How Slop Leaks Into Marketing Teams
The slop-grenade problem is often invisible in marketing telemetry. A typical AI-assisted content flow looks like this:
- Marketer drafts a blog post or campaign brief with AI in 20 minutes (real time saved).
- AI produces a long, well-structured draft.
- Marketer skims, feels it "looks done," sends it to a peer reviewer.
- Reviewer opens it: three sections cite broken links, two statistics are paraphrased from memory rather than sourced, headings read fluently but don't answer the brief.
- Reviewer spends 45 minutes rewriting or marking it up.
- The 40 "saved" minutes disappear; sometimes the net is negative.
Compared with a controlled flow:
- Marketer drafts with AI from a specific brief (20 minutes).
- Marketer runs a quick checklist: claims sourced, links verified, brief intent met (15 minutes).
- Reviewer's job shifts to strategy evaluation rather than basic cleanup (20 minutes).
Total elapsed time is similar, but reviewer effort drops by about half, and the file that lands in the next system (CMS, email tool, ad platform) is closer to final.
This is the operational difference Vivid sees in most teams we assess. AI measures well as a per-draft time saver and badly as a per-collaboration time sink unless the quality gate sits before handoff, not after.
A Practical Quality Gate for AI-Assisted Marketing
At Vivid, the question we ask before any AI-assisted deliverable leaves a desk is simple: "What did a human verify?"
A workable gate has four parts:
- Source check. Every quantitative claim has a primary source, dated and linked. AI-generated statistics are drafts needing verification, not facts.
- Intent check. Does the piece answer the question it claims to answer, in the format the brief specified? A brief for "compare these two platforms objectively" should not produce five paragraphs of feature-by-feature praise.
- Link check. A 30-second click-through is cheaper than a colleague's 15-minute investigation.
- Handoff note. Two sentences at the top: what you wanted the reviewer to verify, and what you already verified. Claiming credit for what you checked is what separates a draft from a slop grenade.
That gate takes 10 to 20 minutes per asset and pays for itself the first time a reviewer doesn't have to rebuild a draft.
The Hidden Cost: Trust, Not Just Time
The workslop data shows a second-order effect worth flagging. Of workers who received workslop, 36% reported wanting to avoid working with the colleague who sent it. Slop doesn't just cost time; it erodes team trust (BetterUp/Stanford study, 2026).
In a marketing team, trust erosion compounds. If a content manager ships AI-generated briefs that editors have to deconstruct every time, editors stop expecting briefs to be useful. They start informal back-channels. The "AI productivity" headline turns into a tax on the collaboration loop.
The Leadership Signal That Matters
Lutke's and Von Ahn's narratives are part of a wider 2026 shift: AI-first CEOs are now talking publicly about AI accountability, not just AI adoption. That's the signal marketing and operations leaders should mirror on their own teams.
A team that hits high AI adoption without a quality gate isn't more productive. It's producing more volume at marginal quality, with the cleanup tax hidden in peer review. The first metric worth tracking is not "how many drafts did AI produce this month?" It's "how many drafts shipped to a reviewer without a basic human check?"
If that number is high, you have slop grenades. The fix is not less AI. The fix is a 15-minute gate.
Read next:
- September 2026 State of Search: Gemini 3.7 Flash Hits AI Mode
- Alibaba's AgentCore and Agentic Cloud: What the Full-Stack AI Strategy Means
- Google's AI Threat Report: Agents Steal Thousands of Credentials in 6 Hours
- AI-Generated Ads Are 14% Worse Than Human-Made: What the HBR Research Actually Says
Wondering where AI is leaking productivity in your marketing team? Contact Vivid for an audit that separates real AI gains from slop-grenade cleanup costs.
