2022 was the year of autocomplete
2023 was the year of chat
2024 was the year of the AI IDE
2025 was the year of coding agents
2026 was the year of agent orchestration
2027 will be the year of outcome engineering
There's a new version of this archetype every few years.
14 years ago you would not believe the number of people who told me that Wix, Squarespace, Weebly, etc.. would make web developers irrelevant when I got into coding.
My advice for anyone looking for a successful long-term career in tech: aggressively move into higher levels of abstraction as quickly as they are made accessible to you, forever.
Going into 2027 if you
• Are using a capable harness
• Understand the problem you're trying to solve
• Are able to clearly break down that problem
You should no longer be managing (or even having to think about) skills, context, verification, /goal, automations, memory, plugins, connectors, subagents, ... other plumbing.
Your harness should be capable enough to manage all of these things for you.
Skills specifically are increasingly unnecessary, counterproductive, and becoming an old-school artifact, only needed for edge cases (and what I'd consider the new version of code smell).
This is great because it eliminates most of the mental overhead we can now use purely focused on problem solving.
With Devin all of this is rolled into one sub:
• Free SWE forever (currently SWE-2, a great model)
• All frontier models (Claude models like Opus and Fable, Gemini models like Neon and Carbon, ChatGPT models like Astra and Sol, Grok)
• The best VMs of any cloud agents
• Devin Bots (agents that own outcomes, not tasks)
• Devin Review
• Devin Security
• Cloud VM optionality (macOS, Linux, Windows)
• Open models like GLM, Kimi, DeepSeek)
• Fusion router (Frontier at ~40% lower cost)
• Devin Oncall
• Desktop, CLI, web w/ iOS and Android available for early preview
• Bring your ChatGPT sub
Sharing because I've been chatting with a lot of new users lately who weren't aware they were already paying for all of this.
Okay, Devin Cloud is highly underrated
I believe only 10% of users really understand its potential
Devin currently offers 4 free SWE-2 concurrent cloud sessions, each with its own VM (8 vCPU, 32 GB RAM), and on top of that it can be macOS or Windows
Just think about that…
On Friday IN DUBAI, we had our @DevinAI Founders DEMO NIGHT!!!
I got to do my own Demo on DEVIN BOTS, showcasing the different bots I have built for my event management workflow, including a reward system that leads to a daily video of my Chief of Staff bot crowning the bot of the day!
We then followed that with 6 demos from founders in Dubai, ranging from early-stage startups to ones who are raising at an 8-figure valuation
I am so happy to see the startup community in Dubai grow, and grateful that so many founders are adopting Devin, hehe. Thank you all for coming!
I have been using Devin just for 2 days since I got the max sub after pitching @gyblos
and I have to say that’s so fucking good. It’s found more bugs than Claude Code and Codex both together.
I always do cross reviews. Devin’s PRs reviews feature is awesome itself
Also the bots. I use the software factory one and it found a lot of little things worth to be fixed
Have to admit that I was skeptical about it before but @cognition cooked, great job!
I’ll be sharing more on my youtube channel, in spanish though
I feel that the progression from "goal" to "outcome" based engineering is the biggest advancement our industry has seen since vibe coding.
For the first time we have proactive, perpetual, self-verifying systems that make it possible. Some examples of what this looks like:
Goal: "Build an Android app."
→ Outcome: "Keep our iOS and Android apps at feature parity on every release."
Goal: "Translate the app into Spanish."
→ Outcome: "Keep the product fully localized in our top 12 languages, in sync with every English copy change."
Goal: "Fix this crash."
→ Outcome: "Keep the crash-free session rate above 99.9% on every release."
Goal: "Optimize the homepage."
→ Outcome: "Keep every page at a Lighthouse score of 95 or higher."
Goal: "Set up backups."
→ Outcome: "Make sure we can always restore production from backup in under an hour, and prove it weekly."
Goal: "Fix the bug from this support ticket."
→ Outcome: "Every support ticket caused by a bug ends in a shipped fix and a reply to the customer."
Goal: "Upgrade our dependencies."
→ Outcome: "Keep every dependency on its latest stable version with CI green."
Goal: "Clean up the dead code."
→ Outcome: "No feature flag, dead code path, or unused endpoint survives more than 30 days."
Goal: "Write a setup guide."
→ Outcome: "A new engineer can always go from fresh clone to running app with one command."
Goal: "Update the docs."
→ Outcome: "Keep the docs accurate to whatever is on main."
Goal: "Refresh the staging data."
→ Outcome: "Keep staging an exact, anonymized mirror of production."
Goal: "Write a competitor comparison page."
→ Outcome: "Keep our competitor comparison page accurate as their pricing and features change."
Goal: "Speed up the API."
→ Outcome: "Get p95 API latency under 200ms and hold it there as traffic grows."
Goal: "Run an accessibility audit."
→ Outcome: "Keep every page of the site WCAG AA compliant, forever."
Goal: "Fix this production error."
→ Outcome: "Every production error gets a root-cause fix and a regression test within an hour of first appearing."
@wojkuli Good feedback! I hope to have something like this early next week. Glad you're appreciating the iOS/macOS environments, I'm a power user and think they are underrated
Recommend using SWE-2 as child for Bots until you really need frontier, we're also working on cost reduction measures for Bots. I've heard SWE-2 is pretty good for design and also feel that way but definitely not Opus 5.5 level, we have new SWE-2 coming soon, curious your thoughts for it when it rolls out!