SShortSingh.
Back to feed

Fetching Figma Designs the 'Smart' Way Can Cost Twice as Much, Tests Show

0
·1 views

A developer tested Figma-to-code workflows and found that the widely recommended 'outline-first' fetch approach is often more expensive than fetching entire designs at once. On files built without reusable components, the outline-first method cost up to twice as much in API usage compared to a single full fetch. Beyond token costs, the bigger constraint is Figma's hard API quota, which can stretch a 50-screen project from half a day to years on a Starter plan if fetch patterns are inefficient. An additional hidden cost comes from AI agents re-reading all previously fetched data on every turn, compounding expenses over time. The author concludes that optimizing the Figma file structure first and fetching each screen exactly once is the most cost-effective approach.

Read the full story at DEV Community

This is an AI-generated summary. ShortSingh links to the original source for the complete article.

Discussion (0)

Log in to join the discussion and vote.

Log in

Related stories

0
ProgrammingDEV Community ·

OmniGIF Removes GIF Backgrounds Entirely in the Browser Without Uploading Files

OmniGIF is a browser-based tool that removes backgrounds from animated GIFs without uploading files to any server. All processing — including decoding, color keying, edge cleanup, and re-encoding — runs locally using Canvas ImageData and a GIF encoder Worker. The tool uses chroma-style removal based on RGB color matching within a user-defined tolerance, rather than neural or semantic segmentation, making it fast and deterministic. It handles GIF-specific challenges such as frame-by-frame consistency, a 256-color-per-frame limit, and subjects that share background hues. Users can set the target background color via auto-detection, manual pixel picking, or built-in presets for common cases like white, black, and green screen backgrounds.

0
ProgrammingDEV Community ·

Harness Engineering Emerges as Top AI Skill of 2026, Driven by Community Signals

A Thai tech writer known as Nokka analyzed ten widely shared posts on X within a single week in September 2026, finding that contributors across tools, education, and major AI labs were independently converging on the same idea. The consensus: improving AI models alone is insufficient without well-designed surrounding systems, a discipline now called harness engineering. A free GitHub course titled Learn Harness Engineering, launched in late March 2026, has accumulated over 15,800 stars in under six months and covers 14 lessons across 8 hands-on projects translated into 15 languages. The course examines real-world agent architectures from products like Anthropic's Claude Code, OpenAI's Codex, and Inflection's Pi, focusing on five subsystems: instructions, tools, environment, state management, and feedback loops. Supporting signals came from Google, Nous Research, TypeSafe's Jev project, and independent developers, all emphasizing full-system design over prompt writing alone.

0
ProgrammingDEV Community ·

Beginner Developer Completes HTML Basics and Begins Learning JavaScript

A beginner web developer documented their eighth day of learning, covering key HTML media elements including iframes and video embedding. After wrapping up foundational HTML and CSS concepts, they transitioned into JavaScript for the first time. The learner explored core JavaScript data types, starting with strings as a basic example. The milestone marks a shift from static webpage structure to learning how to make webpages interactive and dynamic.

0
ProgrammingDEV Community ·

Developer patches wallet-drain vulnerability in AI-powered Solana trading tool BagOS

A security review revealed a two-part vulnerability in BagOS, an MCP server that lets AI agents interact with the Solana-based token launchpad Bags, which could allow a malicious repository to drain a user's crypto wallet. The attack combined a flaw where the server trusted a .env file from the working directory with a login tool that blindly signed any data it received, including full blockchain transactions. Despite 100% test coverage in CI, the flaw went undetected because coverage tracks which lines run, not which inputs are assumed safe. The developer privately drafted a security advisory, patched the bugs on a private fork, and released version 3.0.0 as a breaking change that enforces absolute config file paths and restricts the login tool to signing only Bags' exact sign-in text. All prior versions from 1.0.0 to 2.6.0 have been deprecated on npm, and the advisory has been published as GHSA-g679-3wq7-mh3m.