๐๐ก๐ ๐ญ๐๐ฌ๐ค๐ฌ ๐ญ๐ก๐๐ญ ๐ฎ๐ฌ๐๐ ๐ญ๐จ ๐๐๐ ๐ฎ๐ฉ ๐ญ๐จ 80% ๐จ๐ ๐จ๐ฎ๐ซ ๐ญ๐ข๐ฆ๐ ๐๐ญ ๐ฆ๐ฒ ๐ฉ๐ซ๐๐ฏ๐ข๐จ๐ฎ๐ฌ ๐ฌ๐ญ๐๐ซ๐ญ๐ฎ๐ฉ, ๐ง๐จ๐ฐ ๐ญ๐๐ค๐ 20% ๐๐ญ ๐ ๐ข๐ง๐ - ๐๐๐๐๐ฎ๐ฌ๐ ๐จ๐ ๐ญ๐ก๐ ๐ฐ๐๐ฒ ๐ฐ๐'๐ซ๐ ๐ฎ๐ฌ๐ข๐ง๐ ๐๐.
We've offloaded all our high-cognitive load, limited system-knowledge tasks to AI Agents (for more on that see my previous substack).
When you add it all up, it's taking 1/4 of the time it used to take - because of how much work the AI Agents do for us.
Once the specs are written, the developer involvement is down to 4 steps - of which 3 are just one click each.
Delegate
- the issue (in the web app, Linear or Slack).
- Approve the implementation plan
- Review the code changes, logs and preview (this is where it's more than one click) a. Approve b. Make quick manual revisions c. Prompt Agent to revise code
- Create PR
The Agent is doing everything else ๐๐ ๐๐๐ ๐๐๐.
So all of those UX/UI improvements, minor bugs to squash and simple features to develop - are done by the first coffee break of the day, leaving us to spend time on planning the future of the product and the new challenges we are undertaking.
P.s. A key factor is identifying which tasks are best offloaded to AI. I shared practical advice, key criteria and an AI script with Fine users and my substack subscribers. If youโre interested in reading it, comment below and Iโll send you the link.