Geoff Goodman
@filearts
Followers
1K
Following
5K
Media
161
Statuses
5K
Curious mind. Creator of @plnkrco. Tech lead for extensibility at @auth0 (@okta). Optimistic troll. Never stop experimenting.
Montreal, Canada
Joined April 2011
RT @dschenkelman: if you work at a large company and are deploying your MCP servers to prod, how are you handling auth?. language/fwk sdk?….
0
2
0
Grug is your AI Agent. Optimizing for Grug is probably going to help your human SWEs as well. Their wetware just happens to be more adaptable for now. The thing is that you can have as many Grugs as you want. No hiring and super consistent performance.
Grug is an incredible engineer. He will tackle any project, no matter how boring or repetitive. He will never stop either. He'll even thank you for putting him to work. Problem is, Grug has no hands, no eyes and never remembers anything you told him in past conversations.
0
0
2
Grug will absolutely thrive with the right DX. And Grug will be nothing but a novel distraction with the wrong DX. How are YOU setting your team up to get the most out of Grug?.
0
0
2
You will be equally surprised by his brilliance as you are annoyed by his total inability to unblock himself like the rest of the team. Grug is a mirror that reflects back the true state of the developer experience you have created.
1
0
1
Imagine what you can get done with Grug on your team though!. You'll have to put in the time to empower him. He doesn't even trust himself to run shell commands (and nor should you). But give him the right tools and he will blow you away. Grug is a savant like this.
1
0
1
Grug is an incredible engineer. He will tackle any project, no matter how boring or repetitive. He will never stop either. He'll even thank you for putting him to work. Problem is, Grug has no hands, no eyes and never remembers anything you told him in past conversations.
1
0
1
RT @MarkJCarney: Canada is the greatest nation on earth. Our destiny is to make it greater still.
0
1K
0
This is what I mean by stateful: a model whose own weights evolve over time. A model whose ' is obviated because the model self-improving instead of frozen in time.
When you reframe the problem this way, it really surfaces the stateless nature of LLMs. How and when will we start to see stateful LLMs? Do we even want them?.
1
0
0
When you reframe the problem this way, it really surfaces the stateless nature of LLMs. How and when will we start to see stateful LLMs? Do we even want them?.
+1 for "context engineering" over "prompt engineering". People associate prompts with short task descriptions you'd give an LLM in your day-to-day use. When in every industrial-strength LLM app, context engineering is the delicate art and science of filling the context window.
2
0
2
Pre-commit hook that asks your friendly neighborhood agent to check for errors, omissions, missing doc updates, etc. Overhead or value?.
1
0
1
If my goal was to explore what is possible and inspire new ideas with a quick feedback loop, I certainly would. I wouldn't build those bridges but they would be very valuable input for the bridge I do build.
Would you hire a civil engineer who can "vibe design" 10x more bridges than a regular engineer?.
0
0
1
MCP creates dynamics that are not easily (or cleanly) handled by existing auth RFCs. There are new participants and a complex matrix of potential trust relationships between them (or not). The happy path is probably first party MCP servers for platforms.
Turns out, fixing APIs to specific OAuth servers let all sorts of security concerns be ignored. MCP is the canary in the coal mine for these problems.
0
0
1