Published on
May 30, 2025
At Sierra Ventures, we recently hosted an EPL Council dinner featuring over 30 engineering and product leaders from category-leading companies, including Harness, Drata, Airtable, Square, Instacart, Gusto, Scale AI, and over a dozen others. The conversations were candid and focused, with less hype and more substance. We split the discussion into two tracks: engineering and product.
This post summarizes the most compelling themes from that discussion, encompassing AI code generation, agentic systems, product development, and the impact on go-to-market strategies.
From the Engineering Side: Generation, Oversight, and Real-World Limits
AI Code Generation Is Helpful—But Not Hands-Off
AI is proving valuable for new software projects, with teams reporting that more than 50% of the code can now be auto-generated. However, when it comes to legacy systems, the lift is far lower. Outdated dependencies, complex architectures, and inconsistent patterns all limit AI’s effectiveness. Engineering leaders see AI as a speed enhancer—but not a substitute for system-level knowledge or human accountability.
Front-End Engineering Sees Faster AI ROI
Front-end workflows are where AI is driving immediate returns. Tools that convert Figma designs to code and allow product managers to prototype directly deliver measurable impact. By contrast, backend automation remains challenging due to the complexity of data models, infrastructure logic, and integration layers.
Product Requirements Are the New Bottleneck
AI forces rigor. If the inputs are vague, the outputs degrade fast. Teams are learning that more clearly defined product requirements lead to better AI outcomes. This shift is prompting some organizations to incorporate AI earlier in the development cycle—at the product specification and planning stage—so that requirements can be stress-tested and refined before development commences.
AI Code Reviews Require a Human in the Loop
Engineering teams are experimenting with AI-assisted pull request reviews to streamline workflows. But there's a shared concern: AI reviewing AI-written code without human oversight introduces risk. The most effective setups utilize AI to handle initial reviews, with experienced engineers making the final decisions on code quality and architectural fit.
Bug Triage and Log Analysis Are AI Sweet Spots
AI agents are excelling in areas like bug triage and system log analysis—repetitive, time-consuming tasks that don’t require deep system context. Leaders report increased speed and reduced toil in incident management processes, freeing up engineers for more strategic work without compromising quality.
Engineering Skepticism Around “AI Agents” Remains
There’s growing pushback on overhyped “AI agents” that are little more than glorified loops. Engineering leaders want clarity: how does it work, where has it been deployed, and how does it support developer experience? The consensus: agents will mature, but they’re not yet delivering enterprise-grade outcomes.
Latency Kills Trust—But Feedback Builds It
User experience (UX) is critical for AI adoption. If an AI tool takes more than 20 seconds to respond, users start to disengage. However, when tools provide streaming responses or show visible progress (like intermediate reasoning steps), users remain engaged. UX is emerging as a key competitive lever—just as important as model performance.
From the Product Side: Design, Pricing, and New Metrics for AI
Pricing AI Products Is Still a Puzzle
Product teams are wrestling with AI monetization. Usage-based pricing is becoming common, but it’s far from straightforward. AI-driven features often carry significant infrastructure costs, deliver variable value across users, and add complexity to product portfolios. The most effective pricing strategies combine usage and subscription models grounded in deep user research.
Users Expect Proactive, Not Reactive, AI Products
Customers no longer just want tools that respond to input—they expect intelligence that anticipates needs. As agents become autonomous actors within systems, UI/UX and product design must evolve to support both human workflows and machine-led interactions.
Traditional Product Metrics Don’t Work for Agents
Product teams are realizing that standard engagement metrics (like clicks and sessions) don’t apply when AI agents are doing the work. Instead, teams are trying to redefine success around outcomes: task completion, speed to value, and reduced time-to-resolution.
Final Takeaway: AI Is Delivering Real Value—When Applied Thoughtfully
The most successful engineering and product teams aren’t getting distracted by hype. They’re integrating AI where it provides real leverage—while asking hard questions about performance, accountability, and workflow fit. As AI continues to evolve, these insights are shaping how companies build, ship, and scale smarter products.