TL;DR
|
|
Building your own coding agent sandbox is a trap. "Just run it in a container" turns into credentialed, stateful workloads with auditing and suddenly you are rebuilding a platform team.
Software is an industrial system. Everyone still talks about software like craft. Our CTO argues the real work is a production line where code is just one station.
Enterprise "agentic SDLC" is more than copilot. Autonomy means plan-run-verify-recover loops, and it only works with guardrails that keep it reliable and safe.
Ona updates worth knowing. Our January changelog, Ralph keeps grinding until there is a real outcome, Canvas runs parallel agents, and open source credits help maintainers keep up.
|
Ecosystem watch
Anthropic's January 2026 Economic Index is a power move: publish a real scorecard, define the terms of "impact," and force the debate out of vibes and into numbers.
But metrics can mislead, since usage often reflects distribution, pricing, and compliance friction as much as model quality, and the ecosystem will start gaming whatever you choose to measure.
Everyone thinks "agentic SDLC" is just copilots with better prompts until you realize autonomy means agents have to plan, run, verify, and recover, not just suggest code.
This piece lays out why enterprises get stuck at assistance, and what has to change for real autonomy to work without turning your delivery process into an unreliable science project.
FastMCP 3.0 is trying to grow up. It's no longer just 'here's a tool server,' it's 'here's an app that manages context,' what the agent sees, what it remembers, where data comes from, how it changes per user, and how state carries across sessions.
Ona updates
Everyone still talks about software like craft, but Chris, our CTO, argues the real work is a production line where code is just one station, and it is rarely the bottleneck anymore.
Once AI speeds up "writing," the constraint moves downstream into review, CI, integration, and release, so the teams that win will not generate the most code, they will move ideas through the whole system with the least friction.
|
New agent mode
Most "autonomous" agents still stall the second they hit ambiguity. Ralph is our loop that keeps going, retries intelligently, and drives toward a merged outcome instead of a clever draft.
|
If you maintain or seriously contribute to open source, we want you using the same tooling we use. Ona for Open Source gives you up to $200/month in credits so backlog, reviews, and dependency churn do not become a second job.
When one agent is not enough, Canvas is how you run several in parallel, each in its own sandbox, in a single shared workspace you can steer like a command center. It is built for "do the whole migration" energy, not chat threads and tab juggling.
A few things that should make your week less annoying:
|
|
JetBrains warmup for prebuilds. Enable Prebuilds to download the backend and build indexes ahead of time. Open your project, and code navigation works immediately.
|
|
|
Agent Skills support. You can now teach the agent your team's workflows. Drop a SKILL.md file in .ona/skills/ describing a multi-step process (deploy steps, review checklists, whatever), and the agent picks it up automatically.
|
|
|
TODO grouping in conversations. When the agent works through a multi-step task, items now show up in a collapsible list with progress tracking (e.g., '3/5 done'). Much easier to follow what's happening without scrolling through walls of output.
|
|
|
Claude Opus 4.5 on Ona Cloud. New conversations now use Opus 4.5 by default. The best sign it's working is that you stop thinking about it.
|
Full changelog →
Try Ona, free
You've scrolled this far. That migration isn't going to do itself. Give Ona two minutes and your most dreaded ticket, see what comes back. Start now →
|
Replay events
Lou (Field CTO) and Matt (Head of Engineering) walk through deploying AI engineers across hundreds of repos at once: CVE remediation, migrations, standardization. Live demos, customer stories, and honest takes on what's still hard. Watch →
'The most powerful agent is the one in the hands of an expert engineer.' Patrick from Kingland on what actually works when compliance isn't optional. Watch →
May your PRs merge without the dreaded 'small nit',
Lou
|