DISCUSSION
Is agentic coding like ClaudeCode / Codex the only direction we need?
I’m starting this thread because agentic coding is getting a lot of attention, but I’m not sure it should be the only direction we focus on. The JEV post https://towardsdatascience.com/an-introduction-to-jev/ suggests there may be another useful way of thinking about AI-assisted work, and I suspect we also need a few supporting practices.
What I’m trying to decide
- Should we treat agentic coding as the main path?
- Is JEV a complementary direction, or just a framing for the same problem?
- What other practices should sit alongside these, such as evaluation, human review, or workflow design?
Possible directions
| Direction | What it adds | Open question |
|---|---|---|
| Agentic coding | More autonomy, repeated actions, task completion | When does it become too much autonomy? |
| JEV | A different way of thinking about AI-assisted decisions | How does it change day-to-day practice? |
| Evaluation | More trust in outputs | What should we evaluate for real projects? |
| Human review | Accountability and judgment | Where is the right level of human involvement? |
A simple model
flowchart TD
A["AI-assisted coding"] --> B["Agentic coding"]
A --> C["JEV"]
A --> D["Evaluation"]
A --> E["Human review"]
B --> F["Automation"]
C --> G["Practical judgment"]
D --> H["Trust"]
E --> I["Accountability"]
A quick decision rule
if task is well-scoped and repeatable:
try agentic coding
if task is ambiguous or high-impact:
add JEV-style thinking, evaluation, and human review
Questions for the thread
- Is agentic coding enough on its own?
- Does JEV fill a gap that agentic coding misses?
- What other directions should we add to the list?
- What are we seeing in practice that still feels missing?