DigitalOcean Community Tutorials
Java, Java EE, Android, Python, Web Development Tutorials
- Website
- journaldev.com · all feeds from this site
- Feed URL
https://www.journaldev.com/feed
Read it in your FeedListing reader
Or add to: Feedly Inoreader Raw feed
Latest posts
-
Request Queuing and Latency Under Burst Traffic: What Happens When Your Model Gets Overloaded?
Learn how burst traffic affects inference latency, request queues, and model performance, and explore strategies to manage sudden spikes in demand.
-
How do I monitor and trace AI agents running in Production
AI agent observability test: 5 of 5 DigitalOcean Managed Agents sessions fed one bad tool result ran clean and answered wrong. Log full results, not errors.
-
How to Run Untrusted LLM-Generated Code in an Isolated Environment With MicroVMs
Learn how to sandbox untrusted AI agent code and run LLM-generated code in an isolated environment with containers, gVisor, and DigitalOcean MicroVMs.
-
Sandboxing LLM-Generated Code Needs a Different Threat Model
LLM-generated code fails by accident, not attack. We tested 5 sandbox tiers, from Docker to Firecracker, to show what containment and idle time really cost.
-
DigitalOcean Managed Agents latency: cold start and resume
We timed 2,280 agent sandboxes on DigitalOcean Managed Agents and found our first resume measurement was wrong: polling reported 540 ms, but the session was ready the moment the resume call returned, in 249 ms. The article explains why polling misleads, publishes cold start,…
-
Agents in Production, Part 3: Let Your AI Support Agent Send Replies, Without Giving It the Credential
Take the credential out of an AI support agent with DigitalOcean Action Gateway and let routine replies go out on their own. Two live runs, 335 tickets: about one reply in six went out without a human, and nothing sensitive slipped through.
-
Cold Start Mechanics: How Warm Pools and Model Caching Actually Work
Cold starts slow down scalable AI clouds. Learn where the time goes in LLM inference and how warm pools and model caching cut start-up from minutes to seconds.
-
Spot vs Reserved GPUs:-A Practical Cost Guide for AI Startups
Compare spot, reserved, and on-demand GPU pricing for AI startups, with H100 cost math and when interruptions erase the savings.
-
How to Measure Time to First Token for a Coding Agent: A Reproducible Benchmark on DigitalOcean Inference Engine
How to measure cold and cache-warm time to first token for a coding agent on DigitalOcean Serverless Inference.
-
How to Safely Let AI Agents Call Production APIs using DigitalOcean Action Gateway
DigitalOcean Action Gateway lets AI agents call production APIs safely: tool scope, version checks, and approvals cut harmful episodes from 48 in 100 to zero.
-
Testing Serverless Inference at Scale: Real Performance Results
We tested serverless inference with 1,200+ requests. Real throughput, latency, costs, and concurrency limits. Data for AI startup infrastructure decisions.
-
Agents in Production, Part 2: Build an AI-Native Coding Agent That Turns Tickets into Reviewed Pull Requests with Managed Agents
Build a coding agent on DigitalOcean Managed Agents that turns a ticket into a pull request, with a second agent reviewing it. Two live runs, both approved in under 7 minutes for about $0.35.
-
Agents in Production, Part 1: Build an Overnight AI Support Agent using Managed Agents
Build an AI support agent on DigitalOcean Harness Runtime that triages tickets and drafts replies, with human approval to send. Live run: 99 tickets, zero sent.
-
Deployment Isn't the Bottleneck: Inference Economics Is
Deploying an AI model takes an afternoon. Keeping it affordable is the hard part. Learn how cold starts, GPU utilization, and cost per request decide the bill.
-
Game-Based Benchmarks for LLMs: How Models Behave Beyond the Leaderboard
What game-based LLM benchmarks measure, where they work and where they fall short, how models differ in games like poker and crisis simulations, and how to use those behavioral results alongside capability benchmarks when choosing a model for your agents.
-
Should You Use Spot or On-Demand GPUs for Model Training?
Spot vs on-demand GPUs for model training: measured reclaim and resume costs on MI355X and B300, and the break-even discount at which spot actually saves money.
-
How to Find Real Failures in AI Agents with Signals and Test Them with Evaluations
Signals flags the AI agent conversations that went wrong. It watches only the agents you turn it on for, hides personal details, keeps the data in your region, and never uses it to train AI for anyone else. It found all 6 broken sessions in the test I ran, and cut failed tool…
-
Why does cosine similarity fail for semantic caching? Testing cosine, LLM prompts, and a purpose-built verifier: Jev
We tested cosine similarity, prompted LLMs, and a purpose-built verifier (Jev) on 500 labeled query pairs, and cosine was clearly the least accurate, serving near-misses like "capital of Kenya" for "capital of Jamaica." Jev was comparable to strong small LLMs on accuracy at…
-
Full-Stack Beats Point Solutions — The Infrastructure Argument for Running AI on One Cloud
Discover why full-stack solutions outperform point solutions for AI workloads. Explore the infrastructure case for running AI on a unified cloud platform.
-
What real time-to-first-token looks like on DigitalOcean Serverless Inference
Real time-to-first-token numbers from 380 test calls across six models on DigitalOcean Serverless Inference, including two multi-minute delays on one model.