Sabitlenmiş Tweet
Nate
6.3K posts
Nate
@natebjones
Head of Product, Dad of 2 (3 with the corgi). Tweets about what I care about. Roots in SE Asia 🇮🇩 🇵🇭
Seattle, WA Katılım Ocak 2021
2.7K Takip Edilen4.7K Takipçiler
We already had the capability to automate a lot of the more deterministic workflows decades ago and it always required the messy step of sitting down with the SMEs to capture the process knowledge and then lots of testing to find the edge cases that were not volunteered. People don't understand that this step is still required, only universe of automation candidates expanded.
English
FDEs are the scarce talent layer the market is using because enterprise AI lacks a productized deployment layer.
Someone still has to translate messy internal work into permissions, workflows, exceptions, and ownership for agents.
Most orgs can't define that gap specifically enough to hire for it.
Allie K. Miller@alliekmiller
The most expensive mistake in enterprise AI right now: treating FDEs as your whole transformation plan. Forward deployed engineers (FDEs) are important for custom deployments, but they won’t fix the change management issue most enterprises are facing. It’s likely more the former that Anthropic and OpenAI will continue to prioritize (and hire into the thousands, who knows). Beyond performance and cost, it’s systems integration, ROI, and literal usefulness that drive revenue and stickiness. *However* External FDEs, in my opinion, will not make your company an AI-first company. You can have the sleekest multi-agent orchestrations and still have the majority of your employee base hating AI, avoiding AI, and distrusting leadership decisions on AI. And we already know this because we see this in traditional SaaS too: you can customize the heck out of your Salesforce deployment, but that doesn’t mean your sales team will improve their data hygiene or even attempt to change the way they track and grow with it. Buying a fancier car doesn’t mean you magically learn to drive better overnight. If you’re an enterprise exec and FDEs are sold as the immediate and sole solution to your company transformation woes, walk away. It’s the combination of tech *and* people enablement *and* process reinvention that compounds into actual business outcomes. Large complex enterprises will stall out if they only prioritize the first.
English
I think “computer use” is useful, but I don’t think it is the destination.
The deeper thing is that agents are forcing software back toward the actual structure of work: files, state, tools, memory, permissions, context, and execution.
The screen is just one way into that.
English
The AI security question is moving.
Used to be "can models find bugs." Now it's "can organizations handle the volume of what models find."
Discovery is one part. The hard parts are validation, prioritization, patching, deployment, and code that stays understandable enough to keep getting better.
English
The people getting the most out of stronger models are often not doing magic prompts.
They are being clearer about the work.
What should exist at the end?
What source material matters?
What should be checked?
Where should the model be uncertain?
What requires human judgment?
English
Something AI investors need to be more honest about: the public market isn't a lazy benchmark anymore.
If an LP can buy a liquid basket of AI exposure tomorrow, your private strategy has to justify the risk, the fees, and the illiquidity.
"Better story" isn't an answer.
English
I’m increasingly interested in the difference between an agent that can act and an agent that can preserve context.
Acting is obviously useful.
But continuity is what lets the work get better over time: what was tried, what failed, what changed, what the user cares about, what should not be repeated.
English
@natebjones the parts removed first are the ones with clear inputs and outputs. the rest stays. that is also the part that was always the actual job.
English
The job displacement story I find most plausible is also the least cinematic.
AI does not need to replace a role in one clean motion. It can remove pieces of the work while the title stays the same.
Then the next reorg or budget cycle catches up to what already happened.
English
@henrytdowling Yes, locally orchestrated with excellent memory and presumably local file read/write should beat off-shelf cloud
The fun part is that the hyperscalers are making that distinction blurry on purpose. cloud agents now operate on your local, do read/write, etc.
English
@natebjones do you think the "cloud" part matters more or the "personalization" part? ie would locally orchestrated ai with great memory be > off-the-shelf cloud agents?
English
One thing I’ve changed my mind on: the personal AI computer is not mainly about rejecting the cloud.
Cloud models are still going to matter a lot.
The point is that serious agents need proximity to the materials of work. Files. Apps. Local context. Memory.
The things work is actually made of.
English
The best version of an agent doesn't feel like a person DMing you all day.
It feels like a good operator preparing the ground.
Knows what matters.
Knows what can wait.
Knows what context is missing.
Knows when staying quiet until there's a real decision is the most useful thing it can do.
English
Most of the agentic commerce talk is stuck on where the transaction happens.
The transaction is the easy part.
The harder problem is where the decision gets shaped. If the assistant defines the need, compares options, and remembers preferences, the buy already moved before checkout begins.
English
@DanielDayJewish Well, if an agent is putting my credit card into your checkout, I need to trust the whole stack. Not just your site, but the agent, the handoff, the fraud rules, the refund path. Every layer is a place trust can break.
English
The old internet trained businesses to make themselves legible to search engines.
The next version asks them to become legible to agents.
That sounds like a small wording change, but it reaches into inventory, pricing, refunds, delivery promises, fraud rules, loyalty, and trust.
English
Today I watched Codex use computer use to help me set up Supabase + OpenBrain.
I was overwhelmed. I asked for help. Then I watched the cursor move.
It sounds small until you see it happen on your own computer.
AI is no longer just answering questions. It’s starting to do the work with you.
I’m using OpenBrain to organize 9,000+ exported chats from ChatGPT, Claude, and Gemini — basically three years of business ideas and AI experiments.
Huge thank you to @natebjones for building OpenBrain.
This feels like the start of turning scattered thinking into a real business command center.
English