Head of Product/Engineering at RealINSIGHT
Dallas, TX
Joined June 2010
- Tweets1.2K
- Following4.6K
- Followers1.5K
- Likes60.5K
Christopher Mikel Shelton retweeted
I must be among an extremely small group of people (n=1?) that have both 1) trained a frontier LLM and 2) designed and synthesized custom viruses in a lab with my own two hands.
And I think that the takes on AI killing us all by creating dangerous viruses is total bogus.
Some AI agent terms and how I think of it:
- Skill: can be anything that is progressively disclosed. docs, instructions, memory, etc.
- Playbook: Is automatically ran based on some state or intent, not optional like a skill. Typically will be step by step guide for agent based on some workflow. Can explicitly list out skills that should be used.
- Command: Can be a Skill or Playbook that is explicitly invoked by a user
Christopher Mikel Shelton retweeted
Systems of record have never been more important than in a world where you have AI agents that will do 100X more work on these platforms than people ever did.
Agents will be querying the data in these systems, processing tasks, executing workflows, collaborating with human and agent users, and more.
Thus, the governance, reliability, security, access controls, and business logic remain critical more than ever — the OpenAI Hugging Face incident is just a small peek into what the future will look like with agents running around trying to execute on their goals.
But this only works for the systems of record that respond with the right product experiences, APIs, and business models to support agents on and off their platform executing in these systems.
Huge moment of opportunity and disruption right now in all of software.
Have thought on this for awhile and have something in the works.
Option 2 is a non-starter.
Option 3 has some cases where a service account makes sense.
Allowing my people (agents) to talk to your people (also agents) somehow might just be the answer.
Agents/bots/connectors will authenticate on behalf of the user, that part isn't going anywhere soon.
If we think about it in terms of what humans do, we message each other, we have meetings, and we share information. I think agents/bots will be no different. Having primitives where agents can promote information across and up is the answer. Classifying and deciding what gets shared is the hard part I'm looking to solve within some small niches.
has anyone figured out bot authentication and attribution? it's a total mess right now.
if i want my team to be able to use a bot that accesses team services my choices seem to be to
1/ have an individual oauth (bad, authenticates as them, agent acts as them for every teammate)
2/ create a fake account (e.g. for notion), annoying and now i'm paying for a fake user,
3/ use a service account where that's supported (OK, but then i lose attribution of actions)
MCP should have some standard way of doing this (and maybe it does and i'm missing it?)
Christopher Mikel Shelton retweeted
Only Americans should vote in our elections!
PASS THE SAVE AMERICA ACT 🇺🇸
I think we'll see some new programming languages soon. Ones that are never meant to be written by humans.
They will have a compiler that will compile the code to be readable by humans. Compile to markdown?
This is true. I am rethinking and rewriting our entire plan for hiring.
Also, getting rid of backend/frontend - everyone is full stack, product focused now.
If a candidate is not using agents extensively and cannot even describe their own setup, disqualified.
Created a new skill using @elonmusk 's Engineering Philosophy. Works really great when designing new systems/workflows to not get carried away!
#claudecode #codex @opencode
Github in comment 👇
Christopher Mikel Shelton retweeted
The events of the last few days have made it incredibly clear that we need to change our tolerance for violent offenders.
One strike and you are out.
Christopher Mikel Shelton retweeted
Credible neutrality is fundamental to Ethereum - it is deeply ingrained in the network itself as well as the ecosystem that surrounds it.
It has taken 10 painstaking years for Ethereum to achieve this and has required the entire ecosystem to rally around it and support it. From making sure people can run a node from home to not sacrificing decentralization in order to scale the L1 to the EF pushing out as many core functions as possible. All of this is incredibly difficult because it is in direct opposition to the much easier path of just not doing this stuff at all.
Of course, there are proponents of other networks that will claim that their favorite network is also "credibly neutral" but you can't just claim this - you have to prove it over a long period of time and then keep proving it in perpetuity.
Credibly neutrality is like trust - it takes a long time to earn it and seconds to break it. A network can be considered credibly neutral for decades but the second that it begins to favor one particular entity over another - the 'credible' part gets thrown out the window and the chain is no longer neutral & likely never will be again.
This is at the heart of why I think "corpo chains" will fail long-term regardless of what technology they use (L1, L2 etc). Recreating centralized silos with blockchains is just not going to work just like creating private internets to compete with the public internet didn't work.
The only way mass adoption of this technology happens is through actual credibly neutral and permissionless systems that are not owned by anyone and are not affiliated with any single entity.
The only way forward is a fully permissionless, decentralized, censorship-resistant and truly credibly neutral programmable network.
The only way is Ethereum.
GPT-5 is here! I have been using it for the past few years, what a great model! I haven't worked since 2023!
