Before upgrading to OpenCode 2, check your plugins and integrations
OpenCode 2 changes plugin and integration requirements. Check what carries over, what needs migration and when to wait before upgrading.
Blog
Notes on building reliable AI products.
Sarvam cuts 105B model prices, lowering chatbot running costs
Sarvam’s new 105B rates cut input costs more sharply than output. A worked example shows why chatbot savings depend on usage, not the largest discount.
When Claude’s new workflow beta makes sense for document review
Claude Managed Agents can coordinate large document jobs. How to choose workflows or batching, check review coverage and compare costs per accepted document.
Asana reports 76-fold lower model costs in an AI browsing study
Asana’s browser-agent study shows how workflow changes cut estimated model costs—and why businesses need to price complete, usable results.
OpenAI Agents API Prewarming: Prepare Hosted Environments Before Sessions
How Agents API beta users can prepare hosted environments, verify readiness and attach sessions, with retry, configuration and billing caveats.
Gemini Code Assist Sales End: Renewal Deadlines and Antigravity Migration
Google has ended new Gemini Code Assist sales. Check 2027 renewal deadlines, Antigravity purchasing options, enterprise support and CLI migration steps.
Google Model Armor’s Workspace Update: Configure Screening for Gemini Agents
Audit Model Armor’s Workspace screening: US/EU endpoints, filter versions, agent coverage, blocking, failure policies and skipped scans.
Sophos’s Daybreak MDR Workflow: What to Check Before Automating Response
Sophos’s Daybreak case study puts MDR automation in focus. Examine the reported metrics, customer response modes and evidence needed for a bounded pilot.
Carbon-A’s 566 Million Gene Candidates: What the Evidence Supports
How to evaluate Carbon-A’s gene candidates: distinguish coding tracks, reconstructed genes and experimental support before choosing a research workflow.
LightOnOCR-3: Choosing a Model for OCR, Layout and Chart Extraction
Compare LightOnOCR-3 checkpoints, run the documented local client, and validate layout, chart values and RAG provenance before replacing an OCR pipeline.
Vercel AI SDK WebSocket Chat Transport: When to Switch and How to Resume
When to adopt Vercel AI SDK’s WebSocket chat transport, how full-stream replay works, and the server, cancellation and hosting limits to plan for.
Hugging Face ML Intern: Fine-Tuning Budgets, Baselines and Model Review
Use HuggingChat ML Intern with compute caps, baseline comparisons and model-card checks. A practical guide to budgeting and accepting a custom-model experiment.