Stop the AI Slop: How Job Crafting Saves Quality (and Sanity) in Enterprise
Two years into the generative AI boom, many enterprises are drowning in "AI slop." Think passable drafts, shaky facts, and bland copy that looks productive on a dashboard but burns time in review. The result: a hidden tax on quality and morale.
Leaders keep buying bigger models to fix it. The problem isn't model size. It's how work is designed. When AI is pushed top-down as a universal replacement, you get volume without value.
The real issue: Deployment, not capability
Mandates like "use AI for all first drafts" create rework factories. Employees spend minutes generating and hours repairing. That's not transformation-it's deferred editing.
AI is a jagged tool. It's brilliant at some tasks and brittle at others. Without discretion at the edge-where the real work happens-you get noise.
Job crafting is the fix
Job crafting lets employees reshape tasks, flows, and tool use around their strengths and the realities of their work. In practice: a developer uses AI for boilerplate, not system design. A marketer ideates with AI, but humans own the final copy.
Done right, this turns passive AI compliance into intentional, high-leverage workflows. Quality rises because people decide where AI helps and where it hurts.
Shadow AI is your warning signal
Your people are already crafting their jobs-quietly. The 2024 Microsoft Work Trend Index shows large numbers of employees bring their own AI to work. That's unauthorized experimentation, yes-but also a map of what actually works.
Formalize it. Capture the wins. Reduce the risk. Turn shadow usage into sanctioned advantage.
Microsoft Work Trend Index 2024
Why this matters economically
When workers are treated as AI operators, engagement drops and output flattens into sameness. When they're treated as practitioners who wield AI selectively, quality stabilizes and throughput improves where it matters.
This requires new metrics. Stop rewarding word count and lines of code. Start measuring correctness, customer impact, speed-to-insight, and defect rates on AI-assisted work.
A manager's playbook to formalize job crafting
- 1) Map the work: Break roles into tasks. Label each task as AI-helpful, human-only, or hybrid. Use evidence from pilots, not opinions.
- 2) Create "right to refuse" guardrails: Allow employees to opt out of AI where accuracy, safety, or brand voice is at risk. Require a short rationale and an alternative approach.
- 3) Define AI-in/AI-out standards: Codify where AI may propose drafts, and where only humans make final decisions (e.g., legal, architecture, executive comms).
- 4) Build prompt + pattern libraries: Maintain approved prompts, input formats, and examples of good outputs by task and role. Retire anything that produces slop.
- 5) Set up QA pipelines: For AI-assisted tasks, require source citations, fact checks, test coverage, or peer review. Make "no uncited facts" non-negotiable.
- 6) Secure the stack: Approve tools, data access, and redaction policies. Block sensitive data egress. Offer vetted alternatives so shadow tools aren't needed.
- 7) Appoint workflow owners: Identify practitioners who've built reliable AI workflows. Have them document, teach, and iterate with the team.
- 8) Train by role, not by tool: Focus on "How we do X with AI here," not generic model features. If helpful, explore role-based upskilling options like courses by job that mirror your org's functions.
- 9) Update incentives: Reward error reduction, cycle-time gains, and reuse of proven workflows. Stop celebrating sheer output volume.
- 10) Run tight, time-boxed experiments: 4-6 week sprints with clear hypotheses and metrics. Keep what works. Kill what doesn't. Share results org-wide.
Sector snapshots
Financial services: Let analysts offload data extraction and summarization, while human judgment drives insight, risk calls, and client narratives.
Engineering: Use AI for tests, scaffolding, and refactors. Keep architecture, security decisions, and performance-critical code human-led.
Marketing and comms: AI for ideation and outline variants; humans own final messaging, claims, and brand tone.
Design and product: AI for first-pass storyboards and variations; humans curate, combine, and set the bar for taste and usability.
What to watch
- Hallucinations and drift: Require citations and change logs for AI-assisted outputs.
- Compliance and privacy: Pre-approve data sources and retention rules. Build redaction into workflows.
- Vendor lock-in: Keep prompts and evaluation sets portable. Separate data, logic, and model choice.
- Morale signals: If people are secretly bypassing your stack, your process-not your people-is the problem.
The takeaway for leaders
The AI slop problem is a management problem. Quality improves when the people closest to the work decide where AI is a force multiplier-and where it's a liability.
Stop mandating blanket AI use. Sanction job crafting. Install guardrails, capture what works, and scale it. Your best filter isn't a new model; it's the experienced judgment that knows when to turn the machine off.
Your membership also unlocks:
