AI prompt engineering in 2025: What works and what doesn’t | Sander Schulhoff
- Source
- youtube.com
- Author
- Lenny's Podcast
- Date
Context
Sander Schulhoff, who wrote an early prompt engineering guide and ran a prompt-injection competition, says in a Lenny's Podcast interview with Lenny Rachitsky that prompt engineering still matters because each model release is predicted to end it and then does not. Schulhoff separates conversational use, refining answers over a chat, from product use, where one prompt runs on thousands of inputs and is tuned until fixed.
The example: for a medical coding startup, a first attempt with GPT-4 gave little to no accuracy. Adding a long list of documents that had been coded, each with the reasoning behind its codes, raised accuracy by about 70%, as Schulhoff recalls; the interview does not say whether that is a relative gain or percentage points. It is a single anecdote, but it shows the technique of putting worked examples in the prompt. For the security view, see another interview.
- prompt injection — An attack that hides instructions in content an AI will read — a webpage, email, or document — tricking it into following the attacker instead of the user.
videoHow to measure AI developer productivity in 2025 | Nicole Forsgren
videoInside Devin: The AI engineer that's set to write 50% of its company’s code this year | Scott Wu
videoWhy securing AI is harder than anyone expected and guardrails are failing | HackAPrompt CEO
videoInside OpenAI: 2026 is the year of agents, AI’s biggest bottleneck, and why compute isn’t the issue
Checking sign-in…
Loading comments…

