I started thinking about this while leading an internal Ops platform at a previous operations-intensive tech company. As we automated Ops work, I found the bottleneck to shifting left was was defining the processes our Ops team followed. Eng entered late: first, we'd throw Ops resources at a new challenge; then gradually systematize it; wait for PM attention; document the process; and finally engage engineering. This was necessary since systematized bad processes meant scaled mishandling of issues and engineering rework.
Drafting AI offers a different path: teams begin automating processes organically as they discover them, growing toward full automation as understanding deepens. Human review catches edge cases, handles ambiguous situations, and gives an incremental path towards full automation without rework.
Hopefully this lets internal tools teams shift automation further left.
Enterprises these days have built their own 2 layer guardrail system to work around with Hallucinations and its really effective. I found this blog with proper experimentation on real life use case of Doordash: https://dub.sh/how-doordash-mitigates-hallucinations
I started thinking about this while leading an internal Ops platform at a previous operations-intensive tech company. As we automated Ops work, I found the bottleneck to shifting left was was defining the processes our Ops team followed. Eng entered late: first, we'd throw Ops resources at a new challenge; then gradually systematize it; wait for PM attention; document the process; and finally engage engineering. This was necessary since systematized bad processes meant scaled mishandling of issues and engineering rework.
Drafting AI offers a different path: teams begin automating processes organically as they discover them, growing toward full automation as understanding deepens. Human review catches edge cases, handles ambiguous situations, and gives an incremental path towards full automation without rework.
Hopefully this lets internal tools teams shift automation further left.
I think it's right for this to sit at the browser layer to have a human in the loop.
Enterprises these days have built their own 2 layer guardrail system to work around with Hallucinations and its really effective. I found this blog with proper experimentation on real life use case of Doordash: https://dub.sh/how-doordash-mitigates-hallucinations