Rob Allred

Changing the world through architecture. Exploring what’s next in AI, security, software engineering & government. Purveyor of Odd Things.

[Rob Allred @roballred]

Generational thinking for government IT: plan beyond the current system lifecycle and preserve future options by keeping business data and meaning independent of any one system.

[Rob Allred @roballred]

A government form can be technically correct and still fail the resident.

If people abandon it because questions feel intrusive, confusing, or unsafe, that’s a service-delivery failure.

Completion rate may be one of government’s most useful architecture metrics.

[Rob Allred @roballred]

Interesting security crossover: a technique popularized by AI prompt-injection research is now showing up in conventional phishing.

Security research rarely stays inside the category where it started.

AI security may end up teaching us things about ordinary systems too, not just protecting AI.

https://t.co/oibpygbeY5

[Rob Allred @roballred]

Several models disappear from @GitHub Copilot today.

A useful reminder for enterprise AI architecture: models will change much faster than most enterprise systems.

Designing for model replacement isn’t future-proofing. It may simply be normal lifecycle management now.

[Rob Allred @roballred]

An interesting agent design pattern from @Salesforce: don’t give the model thousands of tools.

Give it Discover → Describe → Dispatch.

As enterprise agent capabilities grow, maybe good architecture isn’t maximizing the tools in context. It’s designing a small, stable interface to a large capability surface.
https://t.co/ptX1We60tJ

[Rob Allred @roballred]

Really interesting look at Uber’s software factory from @udaykiran.

One thing that stood out: as models become more interchangeable, organizational context may become the bigger differentiator.

The model matters. What the organization makes available to it may matter more.

https://t.co/y2RC6yhBlm

[Rob Allred @roballred]

@kritikakodes I’ve never considered html a programming language

[Rob Allred @roballred]

Digital identity looks like a technology problem until you try to implement it statewide.

Then it becomes enterprise architecture: shared capabilities, agency boundaries, governance, funding, trust, and adoption.

The hard part of a digital front door may not be building the door.

[Rob Allred @roballred]

agentic AI could make architecture considerably more operational.
#agenticai

[Rob Allred @roballred]

We’re spending a lot of time discussing what permissions AI agents should have.

The harder architecture question may be what happens after permission is granted.

Can we constrain the environment, attribute every action, observe behavior, intervene, and shut it down?
#agenticai

[Rob Allred @roballred]

AI coding keeps making me wonder whether we’ve had software development backwards.

If agents can increasingly regenerate the implementation, maybe code becomes more disposable while intent, architecture, constraints, and acceptance criteria become the durable artifacts.

[Rob Allred @roballred]

@Henryf1w Not sure I would call wisdom a skill but I think it is uniquely human

[Rob Allred @roballred]

@piotrmiecz I’d like to see agent TEVV include the execution environment itself: identity, tool permissions, network reach, approval gates and provenance. Otherwise we may evaluate the intelligence while missing most of the operational risk.

[Rob Allred @roballred]

@herrmanndigital Once things really consolidate, we will have less choices like with everything else, this will be the good ole days

[Rob Allred @roballred]

@pradeepXkapoor In my opinion AGI goes beyond probabilistic word guessing

[Rob Allred @roballred]

@NickpxJ #humanInTheLead

[Rob Allred @roballred]

@coryalthoff I imagine people said this about assembly when transitioning to higher level languages. I am focusing on architecture

[Rob Allred @roballred]

Suppose you believe you’ve developed the first AGI.

Do you announce it immediately?

Or do you first use it internally to accelerate research and widen your lead?

That might be one of the most consequential incentives in the race toward AGI.
#AI

[Rob Allred @roballred]

@Gaus450 Do you need to understand the code or the architecture?

[Rob Allred @roballred]

@enodrift AI makes disciplined software engineering more important, not less important.