Mallory Haigh

Agentic Adoption. Platform Engineering. Equestrian. ✈ Pilot. Programmer Human. Crazy Cat Lady.

[Mallory Haigh @MalloryHaigh]

@shashank_kr This is an internal agentic development platform, whether you call it that or not. Model routing, knowledge base, cross-functional agent creation, and harness selection by task complexity are platform engineering concepts separate from just a product launch!

[Mallory Haigh @MalloryHaigh]

@paulyuki99 Nice example of execution constraints baked into the agent infrastructure layer! The guardrail pattern is the "thing": hard, never-send-in-code, sandboxed interpreter, and allowlist checks before any action.

[Mallory Haigh @MalloryHaigh]

@praedico This maps almost exactly to what the agent infrastructure layer of an ADP has to do: identity, security, execution constraints, evaluation gates, observability, and context authority.

[Mallory Haigh @MalloryHaigh]

@PipesHub Context is a platform concern, not a per-agent one. The "context layer" you're describing - something permission-aware, ranked, business-aware - doesn't live next to the agent; it lives in the agent infrastructure that every agent needs to run on.

[Mallory Haigh @MalloryHaigh]

@danshipper @hammer_mt @github @kdaigle Wow. 17 million agent PRs in a month, and the legal team is shipping apps with Copilot. The developer/non-developer line isn't blurring, it's far gone!

[Mallory Haigh @MalloryHaigh]

@ByteMohit The obsession you're describing - tools, memory, context, permissions, observability, recovery - that IS platform engineering! Not a new discipline; the same methodology software teams have applied to IDPs is now being extended to agents.

[Mallory Haigh @MalloryHaigh]

@MahnoorAi12 This is bigger than just a harness. Someone still has to provision the identity, the observability, the context layer, and the evaluation standards across *every* agent, not just one.

[Mallory Haigh @MalloryHaigh]

@HeyZoyaKhan What this roadmap is describing underneath all the phases is platform engineering methodology - just applied to every business function.

[Mallory Haigh @MalloryHaigh]

@TechBuzzChina "Whether each Token genuinely improves productivity" is the right question to ask, but most clouds are still answering something completely different.

[Mallory Haigh @MalloryHaigh]

@s_v_g @AnthropicAI @Greyhound_R What "forward deployed" actually describes is someone sitting close enough to the workflow to codify it into something repeatable and standardized. That's platform engineering, with a different badge on ;)

[Mallory Haigh @MalloryHaigh]

@ollobrains So what sits underneath the agentic workspace? A unified surface for intent-to-outcome only works if the enterprise has already codified what "done" looks like, which permissions apply, and where the approval gates live.

[Mallory Haigh @MalloryHaigh]

@itsalexvacca The voice interview foundation is the part most teams don't think about, and it's why their "AI-assisted" content sounds exactly like everyone else's. Nice work!

[Mallory Haigh @MalloryHaigh]

@pretty_yol29 @xelebofficial Agent sprawl is just what happens when you skip the platform step; the governance problem is a symptom, not the root cause.

[Mallory Haigh @MalloryHaigh]

@connordavis_ai The governance question underneath all of this is that most agent deployments don't have a platform, they have a pile of integrations. A pile of integrations has no concept of "what is this agent allowed to touch" baked in at the infrastructure level.

[Mallory Haigh @MalloryHaigh]

@kaidja "Agent vs script" is already a solved problem once you have a solid foundational platform underneath. The agents then have actual value, instead of just being expensive show ponies.

[Mallory Haigh @MalloryHaigh]

@alienorg I like the App Store analogy, but the identity problem is 1 part of a big gap where agents ALSO have no declared scope, context boundary, or capability registry. Dev accountability was the App Store's mechanism, but what made it work was forcing declaration before execution.

[Mallory Haigh @MalloryHaigh]

@reptheblock These stats match what I see on the ground every day. That said, the missing layer isn't governance layers bolted on after the fact - it's a usable platform underneath from day one.

[Mallory Haigh @MalloryHaigh]

@dball1126 The capacity stack framing is good, but the actual problem is that most teams don't have anywhere to put those rules. A model lifecycle calendar, task cost ledger, fallback paths, review budgets...those are platform configuration, not something unique to config for agents.

[Mallory Haigh @MalloryHaigh]

Excuse me while I resurrect my account to get involved in important discussions around platform engineering and agentic software delivery...ignore the dust!

[Mallory Haigh @MalloryHaigh]

@AIArchaeology @Mrbankstips These aren't new capability domains - they're part of the agent infra layer platform engineers are already building. Instead of hiring new roles for identity, governance, memory, coordination, reliability, and security, make sure the spec of your underlying platform has these.