From the Front Lines
In this lesson: fundamentals count for more with AI, planning is where the time goes now, and going deep on one topic is what gets you hired.
Top 3 takeaways
Strong fundamentals count for more with AI
The panel's biggest surprise was that good engineering practice became more valuable in an AI-first setup, even though many people assume the opposite. You need enough understanding to guide the agent and to catch it when it goes off track.
Planning is where the time goes now
The work has moved up the stack from typing toward decision velocity, meaning framing the work, architecture, decomposition, and validation. One panelist now spends 50 to 60 percent of a project's time on planning before any code gets written.
Go deep on one topic so you can speak to it
You can build a beautiful app and still struggle in an interview if you cannot explain the choices behind it. Picking one part of each project to study deeply is what comes across as authentic and lands the role.
A panel with Byron Mackay, Megha, Henry, and Adam
Hosted by Byron Mackay, Director of Learning at Gauntlet AI
Byron Mackay is the Director of Curriculum and Learning at Gauntlet AI, where he keeps the team current on how AI is changing the way software gets built, and he hosts this session. Megha spent close to twenty years in software engineering across finance, trading, enterprise systems, and ed tech, graduated from cohort 4, and now works at Nerdy. Henry was a software engineer for about four years, including time at Chewy, graduated from the same cohort, and started a role at a private-equity-adjacent company in Austin. Adam ran his own startup for five years and then worked on the experimental projects team at Stripe, and he was in cohort 5.
Lesson notes
A written walkthrough of the panel, covering what changed, what surprised them, and the advice they would give.
Before Gauntlet, AI Was Mostly Faster Typing
Before joining Gauntlet, the panelists viewed AI primarily as a productivity tool for writing code faster.
Their experience reflected traditional software development: requirements, design, implementation, testing, and deployment. While AI tools accelerated coding, they hadn't fundamentally changed how their teams planned, built, or shipped software.
The Real Shift Is Decision Velocity
The biggest change wasn't writing code faster—it was making decisions faster.
The value of an engineer increasingly comes from defining problems, making architectural decisions, breaking work into manageable pieces, validating results, and iterating quickly. AI compresses implementation, allowing engineers to spend more time thinking about what should be built rather than typing every line themselves.
You Still Need to Understand the System
AI can accelerate work, but it can't replace understanding.
The panel emphasized that engineers should never let the agent know more than they do. When using AI to work in unfamiliar areas, take the time to understand the architecture, research the technology, and validate the implementation. AI is a force multiplier, not a substitute for engineering judgment.
Use Multiple Models
No single model is best at every task.
Several panelists described moving between Claude, ChatGPT, and other tools throughout the planning process to compare ideas and expose blind spots. They also stressed the importance of experimenting with new tools as the ecosystem changes rapidly.
The workflow matters more than loyalty to any one model.
Planning Is Now Most of the Work
The panel consistently returned to one theme: planning has become the highest-leverage part of software development.
Rather than writing massive prompts, they now invest more time creating specifications, defining requirements, and breaking work into clear phases before handing it to coding agents. Manual testing and verification remain essential because today's AI systems are still imperfect.
What's Overhyped and What's Underrated
Not every AI trend deserves equal attention.
The panel argued that large multi-agent systems are often overcomplicated for everyday engineering work, while technologies like RAG and MCP continue to solve real production problems despite frequent claims that they're becoming obsolete.
They also challenged the idea that AI will eliminate software engineering jobs. Instead, AI is increasing the amount of software teams can build, making engineers who can effectively direct AI even more valuable.
Advice for New Builders
The panel's advice was practical: focus less on chasing the newest tools and more on building real products.
Develop production-ready projects, learn system design, understand evaluations and observability, and study at least one area of every project deeply enough to explain it with confidence. Most importantly, cultivate the habit of learning by building—using AI to explore unfamiliar problems while maintaining ownership of the final result.
FAQ
What changes most about engineering when you work AI-first? +
Do engineering fundamentals still count when AI writes the code? +
How much time should you spend planning versus building? +
Is RAG still relevant? +
Will AI replace software engineers? +
How do you prepare for the Gauntlet cohort? +
Is there an age limit to apply, and what makes an applicant stand out? +
What's next?
Keep building with the rest of Night School, or apply to Gauntlet — twelve weeks of technical intensity with the best AI engineers we can find.