Figma fights back with code, AI, and animation layers

Override Labs created 'Is This Okay?', an AI‑driven consent coach that lets teenage boys privately explore ambiguous sexual scenarios. By banning accounts, cookies, and any "green‑light" verdicts, the product embeds privacy‑by‑design and clinical‑grade safeguards, aiming to stop gender‑based violence before it starts.

Figma’s Dylan Field argues AI will accelerate, not endanger, design tools, positioning the product’s Canvas as the natural AI interface. He reflects on the aborted Adobe takeover, the 2025 IPO surge and subsequent market slide, and how AI integration could revive growth. The take‑away: design platforms that embed AI early will capture the next wave.

Figma’s Config 2026 reveal shows the platform adding code layers, motion, and shader tools to keep its canvas relevant as AI pushes product work into code‑first, agentic environments. The gamble is whether designers can stay in the loop or be bypassed by AI‑driven handoffs, reshaping the design‑to‑development pipeline.

Figma now embeds a native animation timeline directly on the design canvas, letting designers create keyframe motion without switching tools. The timeline supports drag‑and‑drop keyframes, auto‑keying, and comments, streamlining handoff to devs and opening motion to the whole team.

DeepSeek Flash’s cheap, fast, text‑only code model slashes browser‑agent inference from dozens of calls to a single planning call, delivering over a 100× cost reduction. That flips the economics of AI‑agent products, letting builders price usage‑based plans and reclaim the moat from big model labs.

Offensive analytics flips data work from crisis‑driven firefighting to a proactive hunt for hidden growth levers. Teams that shift to this mindset surface high‑performing cohorts, sticky features and new experiment ideas, turning steady metrics into step‑function business gains.

Replyt watches Reddit for problem‑solving conversations, drafts replies or DMs in your voice, and ties each interaction to Stripe revenue. The platform lets founders prove which threads actually convert, eliminating guesswork and keeping accounts from getting banned.

Lenny tested Z.AI’s open-weight GLM-5.2 on four production coding tasks, a code‑base audit, UI redesign, and a 45‑minute autonomous bug‑hunting run, and found it on par with Claude Opus while spending only $3.36 for ~6 M tokens. The cheap, vendor‑independent model proves viable for cost‑conscious teams.

Heron is an open‑source passive eBPF/pcap analyzer that reconstructs every LLM call, tool use, and turn from network traffic, even TLS‑encrypted streams, requiring no SDK or proxy. It gives live latency, error rates and per‑agent metrics, letting engineers debug and secure AI agents at the kernel level.

Ben Thompson spent a week building a personal app using AI‑driven vibe coding and distilled ten concrete takeaways. He shows which parts of the development pipeline AI now handles reliably and where human judgment still saves bugs and design missteps. The insights give product teams a realistic map of today’s AI‑assisted coding limits.

CtxGov gives developers a read‑only CLI to surface hidden system prompts, memory state, and context an AI coding agent inherits before it runs. By exposing these instructions, teams can catch risky directives early and enforce safer agent workflows without altering code.

Topos introduces structural code‑quality metrics that evaluate AI‑generated diffs by analyzing control‑flow, module, and data‑flow graphs. By turning nebulous “clean up this code” requests into concrete, measurable targets, it cuts review fatigue and prevents hidden duplication when agents flood repos with machine‑speed changes.