Developer scraps misleading 68% cost-saving claim from open-source AI tool ORA

·1 views

A developer has released ORA, an open-source Go binary that breaks tasks into subtasks and routes each to the most cost-effective AI model available, compatible with tools like Claude Code, Codex, and Cursor. The original README prominently featured a claim that ORA was 68% cheaper than using a flagship model alone, but the figure was based on a hypothetical worked example rather than actual measurements. The developer acknowledged the number was misleading because it ignored that cheaper models often use more tokens and require retries, making raw cost-per-token comparisons unreliable. Rather than refine the metric, the developer deleted it entirely and replaced it with a straightforward display of real token usage multiplied by actual per-model pricing. The move reflects a broader critique of AI tooling benchmarks that compare real runs against imagined baselines chosen by the vendor making the announcement.

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)

How 'Act Like an Owner' Culture Often Leaves Employees With Nothing

Many tech companies tell new hires they are owners by granting equity, but the fine print frequently tells a different story. Standard agreements typically include a four-year vesting schedule, a one-year cliff, and a 90-day window to purchase shares after leaving. Most departing employees cannot afford to exercise their options within that window, effectively forfeiting the stock they were promised. This structure is not an exception or oversight — it is the industry norm. Critics argue the 'ownership' framing is largely performative, designed to motivate workers without transferring meaningful financial stake.

0 comments Read more at DEV Community

ProgrammingDEV Community ·

Three Gorges Dam Ship Lift Moves 15,500 Tonnes Using Counterweight Balance

The Three Gorges Dam on China's Yangtze River houses the world's largest ship lift, designed to vertically transport vessels across a 113-metre elevation difference. Rather than using traditional step-by-step lock systems that take hours, engineers chose to move ships in a single vertical journey inside a water-filled steel basin. The combined weight of the ship, water, and basin reaches approximately 15,500 tonnes, which is nearly offset by a counterweight system so that motors only handle the small remaining imbalance. A rack-and-pinion mechanism ensures all four corners of the basin rise at identical speeds, preventing water from sloshing, while a nut-and-screw safety system instantly locks the basin if a fault occurs. The dam operates two separate transit solutions side by side — the large ship lift for vessels up to 3,000 tonnes and a smaller express channel for lighter boats — completing each vertical crossing in roughly 40 minutes.

0 comments Read more at DEV Community

ProgrammingDEV Community ·

Developer finds critical agentic workflow bugs only visible at runtime, not in code

A developer building ArDD, a structured AI coding tool that breaks projects into plans and tasks for Claude Code agents, discovered significant flaws only after running live smoke tests rather than relying on code review. A key bug revealed that delegated git worktrees branched from the remote tracking branch instead of local HEAD, meaning locally committed state was invisible to subagents. A separate, unexplained issue caused git worktrees to corrupt the primary checkout's configuration by flipping core.bare to true. The developer responded with a mandatory alignment script for subagents and a coordinator-level tripwire to detect the config corruption, while acknowledging neither was a true fix. The experience prompted a broader architectural redesign so that coordinating state only lands on the main branch alongside the code it describes, ensuring failed runs leave no partial state behind.

0 comments Read more at DEV Community

ProgrammingDEV Community ·

Python Tool Converts Stack Traces Into Structured Debugging JSON via AI

A new open-source Python example published on DEV Community transforms raw stack traces into structured JSON debugging output using Telnyx AI Inference. The Flask-based application accepts a stack trace along with optional language and runtime context via a POST endpoint, then returns fields such as root cause, severity, confidence score, and a suggested fix. The structured response format is designed to integrate with developer workflows, enabling use cases like CI failure alerts, Slack notifications, and incident triage dashboards. Recent analyses are stored in memory and can be retrieved by ID through a dedicated GET endpoint. The project is available on GitHub under the Telnyx code examples repository and requires a Telnyx API key to run.

0 comments Read more at DEV Community