Landscape and wildlife photographer. Software engineer. Avid reader.
Header photo: Bowman Lake in Glacier National Park at sunset in late summer.
#wildlifephotography #landscapephotography #photography
| Photography | https://www.weshargrove.com |
| Micro.blog | https://wes.micro.blog |
| GitHub | https://github.com/whargrove |
@snazzyq Maybe, but what if the point is not success per se (as an independent product line) but to test the waters to see what sticks?
Probably the demise of the Plus was baked in to the success of the Pro Max and the better margins associated with the latter.
The risk is real. AI is inherently untrustworthy. Spend a lot of time with the prompt and keep the problem space small for maximum success.
Alternatively, find out ways to use “creative” hallucinations to explore multiple paths and then use human-in-the-loop patterns to guide the AI in a direction that is aligned with your goals.
https://mastodon.social/@rands/115418730710158376
@christianselig @finnvoorhees I agree with both positions. I’ve found the maximally effective approach is to (1) use git worktree to isolate agents, (2) spend a more time in planning mode to spec/scope the work, (3) use bd (https://github.com/steveyegge/beads) to offload the tasks to disk (plan survives compaction), and (4) have the agent stack branches for the work so that it can be reviewed by human (that’s me) in atomic chunks.
Let the agent go wild after scoping and then review / test before merge. 🍰
