AI in Software Development 2026: Portable Skills, ROI, and Fighting ‘AI Workslop’
AI changed the way we write software in 2025. It happened fast. Teams adopted tools and workflows overnight. The hard questions followed. How do we measure return on investment? How do we keep quality? How do we stop AI from producing hollow output that wastes time?
I’m Ermenildo V Castro. Call me Ernie. I worked at Google, Microsoft, Amazon, and Zulily. I live in Seattle. I write code and stories. I also watch trends. I will tell you what matters for 2026.
What are "Skills" and why portability matters
Skills are repeatable workflows you can teach an AI. Anthropic called theirs "Skills". They made them an open standard. That matters.
If you teach one tool a workflow, you should not be trapped. A portable skill works across Claude and other AI platforms. Portability keeps your work safe. It keeps teams honest.
I once taught a coding assistant a build checklist at Google. The checklist lived in one tool. When the tool changed, the checklist was gone. We lost hours. This is why standards matter.
Anthropic and the push for open standards
Anthropic moved Skills toward an open standard. That’s good for engineering. It lowers lock-in. It lets teams share best practices. It lets skills evolve with the platform.
Open standards are boring on the surface. They are heroic in practice. They let you move when the market moves.
Beware the "AI Workslop"
I call bad AI output "AI workslop." It looks like work but adds no value. It’s polished fluff. It wastes time. It gives false confidence.
Teams hand off tasks to AI and treat output as final. That is a mistake. AI can draft code, tests, docs, designs. But drafts must be reviewed. Measure quality, not just speed.
A short memory from my Amazon days: a subsystem delivered a long, neat spec generated by an AI. We assumed it was correct. It passed basic tests and then failed in production. The spec sounded great. It hid a missing edge case. We learned to treat AI output as a starting point.
Measuring AI ROI in engineering
ROI is a simple question with a messy answer. You isolate the AI variable. You measure cycle time, defect rate, and developer focus.
Practical metrics:
- Time saved on tasks that are repeatable.
- Reduction in bug regression rates.
- Developer hours spent on meaningful work vs. rework.
Measure before and after. Use the same task, same team, same deadlines. You will learn more quickly.
Predictions for AI in 2026
- Skills portability becomes a hiring and procurement factor. Vendors that support standards win. Teams will ask for portable skill formats.
- Tools will focus on verifiable output. We will see more linters, test harnesses, and guardrails for AI outputs.
- Rise of AI policy and stewardship roles. Teams will hire engineers to own AI behavior and quality.
- AI will move from assistant to teammate. AI will write drafts and tests. Humans will remain final arbiters.
- More AI-generated noise. Better filters. AI workslop grows. So does the toolset to spot it.
How to prepare your team
- Treat skills as code. Version them. Test them. Review them.
- Make skills portable. Export and import across platforms.
- Add simple verification. Unit tests. Smoke tests. Lint rules.
- Track ROI with a few clear metrics. Start small.
- Train your team to review AI output critically.
Anecdote: small team, big win
At Zulily I led a small team building an A/B experiment platform. We used AI for test scaffolding. We defined the workflow as a skill. We kept it portable. When the vendor changed, we moved the skill in a day. The feature stayed live. The team stayed calm. That day I felt like a steward. Simple faith in good process saved us time. 🙏 — Ernie
Final word
AI will be part of every engineering life in 2026. It will be a tool that saves time or a trick that wastes it. The difference is discipline.
Make skills portable. Measure the work. Watch for AI workslop. Treat AI output as a draft. Build guardrails and tests. Do the hard work first. The rest follows.
— Ermenildo Valdez Castro, Jr. (Ernie) 😊