Latest Tools and AI Agents for Software Engineers
Stay up to date
Stay up to date on the latest dev tools / AI coding tools.
More stories
The Five Stages of Grief When Your AI Agent Asks for Permission to Run `rm -rf`
A satirical field guide to the five emotional stages triggered by an agent’s polite request to recursively erase something it describes as “temporary.” The joke is that the permission dialog is working exactly as designed.
Read moreWrite Project Rules Your Coding Agent Can Actually Follow
Keep agent rules short, scoped to the code they govern, and tied to commands the agent can run. Start with AGENTS.md as the portable baseline, then add tool-specific files only where their scoping features solve a real problem.
Read moreHow to Run Cline Fully Local With Ollama or LM Studio
Cline can run against a model on your own machine with no cloud-model API key: point it at Ollama on port 11434 or LM Studio on port 1234. The setup is short; getting useful agent behavior depends much more on model size, context, and tool-calling reliability than on the dropdown you select.
Read moreContinue a Claude Code or Codex Task in Cline With an Open-Weight Model
Cline Desktop can import local Claude Code and Codex histories, then resume the task using a different configured provider and model. Treat it as a deliberate handoff: verify the working tree, let Cline summarize the foreign transcript, and make the open-weight model prove it understands the current state before it edits.
Read moreCline Desktop’s Parallel Sessions and Scheduled Tasks Change What You Delegate
Cline Desktop’s new parallel execution makes a coordinator-and-specialists workflow more practical, while scheduled tasks turn a few read-only maintenance checks into durable local automations. Neither feature creates safe isolation by itself: use separate worktrees for competing edits, and treat unattended schedules as production automation.
Read moreSet Approval Rules by Blast Radius, Not by Tool Name
A usable coding-agent policy auto-allows cheap, reversible work and stops at state changes: credentials, network access, deployments, protected files, and writes outside the task directory. Start with a read-only baseline, then grant the smallest command and filesystem surface that lets a real task finish.
Read moreStartup Pivots to Self-Hosting LLMs, Discovers It Is Now a GPU Rental Company
A strategic guide to the moment a team replaces an API bill with drivers, node labels, reservation tables, and a deeply personal relationship with idle accelerators.
Read moreBest Cheap LLM API for Coding Agents: 2026 Pick
Gemini 3.7 Flash is the best cheap LLM API for a coding agent’s default model in 2026: it is explicitly positioned for everyday coding and agentic tool use, with paid pricing of $0.75 per million input tokens and $3.75 per million output tokens. Use a still-cheaper general-purpose model only for constrained sub-tasks, not as the model that owns a multi-file change.
Read moreHow to Control AI Coding Costs Without Slowing Your Team Down
Don’t control AI coding spend by making developers ask permission before every agent run. Put hard limits around the expensive failure modes, attribute usage to work that matters, and make the cheaper path the default.
Read moreQuantization for Developers: What Q4, Q8, and FP8 Mean for Code Quality
Use Q4 when it is the model that fits comfortably on the machine you actually develop on; move to Q8 when the agent’s missed details cost more time than the extra memory. FP8 is a different decision: it is usually a GPU/backend execution format, not a local-model quality tier between Q4 and Q8.
Read moreBest AI Coding Agent for AWS Bedrock, Vertex AI & Azure OpenAI Users
For teams already committed to AWS Bedrock, Google Vertex AI, or Azure OpenAI, the right coding agent is the one that keeps cloud identity, regions, endpoints, and model access intact. Here is what to test before you standardize on one.
Read moreDeveloper Gives Coding Agent a Cron Job, Now Receives Pull Requests at 3 A.M. Like a Real Coworker
A satirical field report from the point at which scheduled agent runs stop being automation and become a colleague who respects neither sleep nor issue triage.
Read moreThe AI Model Price War: Choose Models Without Chasing Every Price Cut
Don’t switch your coding model every time a pricing page changes. Set a small, versioned model roster, measure cost per completed task, and make price cuts a routing change—not a new engineering project.
Read moreHow to Use Local Models With Cline for Privacy-Sensitive Codebases
Run Cline against a local Ollama server when code and prompts must stay on a controlled machine, but treat it as a narrower inference boundary—not a complete security model. The setup is straightforward; keeping context, tools, task history, and network access under control is the real work.
Read moreLocal Developer Declares Todo App Too Proprietary for the Cloud
A developer protects a three-table task tracker from remote inference, external backups, and the possibility that someone might discover the business logic behind an unchecked box.
Read moreOpen Source vs. Open Weight: What Model Licenses Actually Allow
“Open weight” tells you that you can get the parameters; it does not tell you what you can ship, fine-tune, host, or call your derivative. For those decisions, ignore the label and read the license, its attached use policy, and the model card as one package.
Read moreAI Agent Refactors 40 Files Perfectly, Senior Engineer Rejects PR Over a Variable Name
A practical postmortem on the only part of an AI-generated 40-file refactor that still required human judgment: deciding whether `items` meant the right kind of items.
Read morePlan Mode vs. Act Mode: Let the Agent Think When the Cost of Being Wrong Is High
Use plan mode before a coding agent touches an unfamiliar subsystem, a public contract, or a migration. Skip it for a localized change when you can already name the files, tests, and acceptable diff.
Read moreOpen-Source AI Coding Agent for VS Code: Best in 2026
The best open-source AI coding agent for VS Code is the one that lets you control model choice, tool permissions, and repository-level instructions without turning normal edits into a configuration project. This guide compares the live options, explains where they break down, and gives a practical default for 2026.
Read moreSelf-Hosted AI Coding Agent for Enterprise Security: Best Pick
OpenHands Enterprise is the best self-hosted AI coding agent for enterprise security when you need a multi-user service with SSO, role-based access, isolated agent containers, and private-cloud or on-premises deployment. Its open-source Agent Canvas is useful for a tightly scoped pilot, but its own deployment documentation says it is single-tenant and lacks built-in user authentication and tenant isolation.
Read moreThe 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 moreFrom IDE Extension to Always-On Service: How Coding Agents Are Changing in 2026
Coding agents are moving out of the editor and into scheduled, event-driven workflows. Treat them less like fast autocomplete and more like a service account that needs a narrow job, bounded permissions, observability, and a review queue.
Read moreHow to Roll Out an AI Coding Agent Without Turning It Into Shadow Infrastructure
Start with a small, measurable cohort and the same branch protections you already trust; don’t start by buying seats for every developer. A useful rollout makes the agent easy to try, hard to misuse, and straightforward to turn off when a repository or workflow is a bad fit.
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 more