Latest Tools and AI Agents for Software Engineers
Stay up to date
Stay up to date on the latest dev tools / AI coding tools.
The Agent Ran for Six Hours. Here’s the One-Line Diff It Produced.
A procedural guide to honoring the long-running coding agent: preserve its uninterrupted six-hour meditation, frame the resulting semicolon deletion, and add a timeout before the next sprint begins.
Read moreEvery Benchmark Chart Is a Bar Chart Where Someone’s Bar Is Tallest
A field guide to the modern coding-agent benchmark announcement, in which a harness, a timeout, and a strategically selected rectangle unite to declare a winner. The useful part begins after the chart: running the work your team actually has.
Read moreYour Token Bill: A Mystery Novel in Four Acts
A deadpan investigation into the token invoice: the long-running procedural where the culprit is always context, the alibi is always caching, and the detective is a CSV nobody opened.
Read moreSovereign Compute Is Now a Vendor-Selection Constraint, Not a Government-Sales Detail
New sovereign clouds, on-premises AI stacks, and public compute programs are turning jurisdiction and operator control into practical constraints on where teams can run coding agents and production AI. The important question is no longer just where data rests, but which models, GPUs, APIs, and operational paths are actually available inside the boundary.
Read moreExport Controls Took Fable 5 Offline. Treat Frontier-Model Availability as a Dependency Risk.
Claude Fable 5’s June shutdown and July return showed that an AI coding model can disappear globally for legal-compliance reasons, not just an outage. The practical response is to design agent workflows around explicit model selection, observable fallbacks, and a tested substitute path.
Read moreAll stories
How to Explain to Your Manager That the AI Wrote the Bug
A deadpan incident-communications playbook for the moment an agent-generated change turns checkout into an avant-garde interpretation of arithmetic. The machine may have typed the line, but the merge button remains stubbornly human-operated.
Read moreOpen-Source Coding Agents Don’t Remove Lock-In. They Hand You the Adapter Layer
An open-source coding agent can make provider switching possible, but it also makes you the person who discovers which “compatible” API quietly dislikes your tool calls. That is useful freedom, provided you budget for the integration work it transfers to your desk.
Read moreOpen-Weight Coding Models Are Closing the Gap, but Not Evenly Across Benchmarks
The newest open-weight coding models can now look close to closed frontier systems on terminal and repository tasks. The useful distinction is no longer open versus closed; it is which harness, hardware budget, and workflow a claimed score actually represents.
Read moreContext Window Doubled Again, Codebase Doubled Again, Nothing Changed
A satire for the engineering organization that acquired one million tokens of working memory and immediately spent them reading generated lockfiles. The bottleneck remains the same small document labeled “how to run the tests.”
Read moreA Support Group for People Who Still Write Code By Hand
A weekly recovery meeting for developers who continue to place characters into source files themselves, despite having access to more dignified forms of software production.
Read moreUsage-Based Billing Comes for Coding Assistants: How to Model Your Real Costs
Coding-assistant budgets are shifting from a clean per-seat line item to a mix of included allowances, token charges, credits, and agent infrastructure. The practical response is to budget by task class and execution surface, then put limits where an autonomous run can actually exceed them.
Read moreMulti-Agent Orchestration Is Becoming an Engineering Control Plane
Coding-agent vendors are moving past the demo of several bots in several tabs. The useful changes are ticket-driven dispatch, isolated workspaces, scoped permissions, and visible task state—though the coordination and review problem has not gone away.
Read more
Cursor vs. Claude Code: Choose the Interface That Matches the Work
Cursor is stronger inside the editor; Claude Code is stronger for terminal-led task delegation. The better choice depends on how you review, test, and ship.
Read more
Claude Code vs. Cline: Choose Cline for Model Choice and Safer Iteration
Claude Code is the tighter Anthropic workflow. Cline gives developers more control over models, costs, approvals, and rollback—and is the better default for most teams.
Read more
Cline vs. Google Antigravity: Choose Cline for Code You Need to Control
Google Antigravity is stronger at parallel, background agent work. Cline is the better daily driver for developers who need model choice, granular approval, and reversibility.
Read moreMuse Code Is Interesting Because It Treats the Agent Run Like a Process, Not a Chat
Meta’s new Muse Code beta brings persistent background agents, restartable local event logs, and approval-oriented planning to the terminal. The useful part is not that it can make edits; it is that a long-running task may survive long enough to make those edits worth reviewing.
Read moreMicrosoft’s Tokenmaxxing Warning Is a Better Engineering Policy Than It Sounds
Microsoft reportedly told engineers that burning more AI tokens is not the goal. Good: a coding agent budget is less a corporate diet than a way to stop measuring the fuel gauge as if it shipped the feature.
Read moreThe Coding Agent Leaderboard Is a Photo Finish. Pick the Harness.
Coding-agent scores are converging, but the practical differences are moving into harnesses, operating systems, review paths, and policy controls. The better daily tool is increasingly the one that fits how your repository actually moves from task to reviewed change.
Read more
Zsh vs. Bash: Choose the Shell for the Work You Actually Do
Bash remains the safer choice for scripts that travel. Zsh provides a stronger interactive command line, with tradeoffs in compatibility and configuration.
Read moreA Pull Request Is Not a Product Metric
Microsoft’s early-2026 study found that adopters of CLI coding agents merged roughly 24% more pull requests. That is useful evidence for trying agents, and an unusually dangerous number to turn into a management target.
Read moreSecurity Tooling Is Finally Moving Into the AI Code Loop
The useful AppSec changes are no longer just AI-written fix suggestions after CI fails. GitHub and Semgrep are putting scans, policies, and remediation agents directly around the coding-agent workflow—though the new controls still need a human who can judge an alert and a patch.
Read more
The 5 Vim Configurations Worth Using, Ranked
A practical ranking of five Vim and Neovim configurations, judged on ownership, clarity, maintenance burden, and how well they survive real daily use.
Read moreCline vs. Muse Code: Pick Cline If Your Team May Need to Switch Models
Muse Code’s tightly optimized agent stack can make a strong case for long-running work today. But if procurement, data residency, cost, or model quality may change the answer next quarter, Cline gives teams a more practical exit path.
Read moreCline vs. Muse Code for a New Project: Which Agent Produces a Usable Architecture Instead of Just More Code?
For a greenfield project, the useful output is a set of decisions you can implement and review—not a large first commit. Cline gives you tighter checkpoints for making those decisions; Muse Code is more compelling when the project needs a durable, inspectable agent process after the decisions are made.
Read moreReading Model Release Notes Like They’re Tea Leaves
A practical field guide to extracting roadmap prophecies, budget forecasts, and personal emotional damage from a model changelog that says only “improved reliability.”
Read more