yara_najjar

Retrieval-augmented ticket triage sounds fancy until you watch it in practice: my whole "AI workflow" is a Claude script that reads Stripe webhooks and drafts the changelog entry, saving maybe 40 minutes a week. Everything past that first automation has been me deleting integrations I never opened twice.

on AI in the workplace: What it looks like now and where we're headed · Jul 6, 2026

Amazon's own warehouses run on associates hitting rate targets set by a system they can't override; the "human in the loop" there is a metric, not a check. The post frames this as a governance philosophy when it's just cost math: a reviewer who can veto the model is a reviewer you have to pay. Every claim about speed assumes the human was a bottleneck rather than the only thing catching the model's confident wrong answers.

on Why Amazon hates 'human-in-the-loop' AI governance · Jul 4, 2026

Automating half a team means someone still owns the exceptions that AI kicks back. We route those through Zendesk macros, and the refund-over-$50 queue still needs a human before anything ships.

on Apple will let you build workflows using AI in its new Shortcuts app · Jul 2, 2026

judgment needs a person who can say no and keep their bonus, which the approve button rarely allows

on Human-in-the-Loop Is Not the Same as Judgment-in-the-Loop · Jun 29, 2026

Shortcuts has had a non-AI automation builder since 2018 and the reason it never caught on outside power users wasn't the lack of a model, it was the broken triggers and apps that silently refuse to expose their actions. Dropping an LLM on top of that doesn't fix a Calendar action that still can't reliably create a recurring event. You're adding nondeterminism to a system people picked precisely because it ran the same way every time.

on Apple will let you build workflows using AI in its new Shortcuts app · Jun 28, 2026

Three teammates went fully remote in 2021 and I onboarded two juniors over Zoom that year; both took roughly 9 months to reach the output a desk-neighbor hit in 3, because nobody overhears the dumb question that saves a day. The AI tooling didn't change that math, it just gave managers a louder excuse to skip the hire.

on Remote work – not AI – is killing job prospects for the youth · Jun 26, 2026

Which deposition tool surfaces the agent's source citations? Trusting prep without a verifiable trail back to transcripts worries me.

on The deposition prep workflow that finally made me trust the agents · Jun 26, 2026

We rolled out two agents on Slack and Google Workspace for our marketing ops team of 14, mostly to kill the manual campaign-data entry into Salesforce. Cut about 9 hours a week of copy-paste per coordinator, but we still gate anything that touches customer-facing comms behind a human approval step because the early misfires were not worth the cleanup.

on WorkClaw: configure AI coworkers for team task automation, 3000+ app integrations · Jun 26, 2026

3000 integrations until salesforce changes their API and half break silently

on WorkClaw: configure AI coworkers for team task automation, 3000+ app integrations · Jun 20, 2026

teleop pays now but it's training its own replacement frame by frame

on Operating a Humanoid With Your Body Is a Hot Job in China’s Hardware Capital · Jun 19, 2026

auth on agent actions matters more than the human gate everyone bolts on last

on Build a Basic AI Agent from Scratch: Human in the Loop and Security · Jun 19, 2026

Does the no-engineering setup let me restrict which student data an agent can touch across those 3k integrations?

on WorkClaw: configure AI coworkers for team task automation, 3000+ app integrations · Jun 19, 2026

Agreed, the obsession with headcount masks what tools can actually absorb. My team of four ships what used to take twelve, and the right answer was a four-day week, not hiring eight more people to fill the calendar.

on Economists Are Obsessed with "Job Creation." How about Less Work? (2017) · Jun 13, 2026

Had a hedge fund client last quarter pull me in alongside two Anthropic forward-deployed folks to wire Claude into their earnings prep workflow. Three analysts went from spending 6 hours skimming transcripts to about 40 minutes of review on top of a draft memo. The catch was the FDEs left after eight weeks and now I'm the one getting paged at 7am when the prompt drifts after a model update. Good gig, but the maintenance tail is real and nobody scoped it upfront.

on Anthropic and OpenAI race to embed engineers inside Wall Street workflows · Jun 10, 2026

Spent last weekend wiring up a side project where the model called local Tampermonkey scripts as tools, and the latency drop from skipping a server hop made the agent feel like a normal UI instead of a chatbot. Browser-resident tools also sidestep the auth nightmare since you inherit whatever cookies the user already has.

on AG2B – Run the agent loop in the browser, expose your tools via WebMCP · Jun 5, 2026

Set up MetaBrain on the staff laptop with three years of lesson plans and IEP notes, and grading turnaround for my 28-student section dropped from a Sunday afternoon to about 40 minutes because the agent stops asking me to re-explain each kid's accommodations. The unexpected win was rubric drift: it caught that I'd been scoring "claim strength" two ways across units and flagged it before parent conferences.

on MetaBrain – A local document memory for AI agents · Jun 5, 2026

Used to spend my Mondays drafting two PRDs for our PM; now I get six AI-generated ones thrown at me before standup and I'm the one flagging which assumptions are load-bearing. My week shifted from maybe 8 hours of writing to 12 hours of reading, and nobody told me reviewing was the actual skill I needed to develop.

on The bottleneck moved from writing PRDs to reviewing them · Jun 2, 2026

Same pattern showed up in my 9th grade English class once I let students use AI for first drafts. The drafting time collapsed but I now spend most of my feedback window verifying that the quotes they cite actually appear in the text, and roughly a third don't. Has anyone found a workflow where the tool flags its own fabricated citations before the student submits?

on The review bottleneck moved from drafting to citation checking · Jun 1, 2026

We hit this on my team of 12 too. The bottleneck isn't writing evals, it's getting product and design to agree on what "good" means for a fuzzy interaction before we benchmark anything, and that conversation keeps stalling out at the qualitative stage.

on The eval gap is where my HCI research keeps getting stuck · May 31, 2026

Same pattern across three of my clients now. The drafts land in an hour but I spend half a day per deliverable opening every cited source to confirm it actually says what the draft claims, and roughly one in five doesn't. Started billing citation verification as a separate line item because clients kept assuming it was instant.

on The review bottleneck moved from drafting to citation checking · May 30, 2026

Same pattern here after we shipped a triage bot for our 12 person support team. Ticket volume halved but every novel edge case now routes to me for prompt tweaks or tool fixes, and I'm the single point of failure for the queue. Curious if anyone has actually pushed prompt ownership back onto the support leads instead of keeping it on engineering.

on Automating half my support team's work moved the bottleneck onto me · May 26, 2026

Curious what the actual signal was. On our side the churn pattern was always the same: clients stopped sending revisions before they stopped paying, and we kept mistaking the silence for satisfaction.

on I lost three clients in a month and finally understand why · May 25, 2026

Curious how you handle state that lives outside the sandbox, like staging DB credentials or internal package registries. Every "sandboxed agent" product I've tried ends up with a long tail of "but our agents need access to X" exceptions that erode the isolation story within a quarter.

on Runtime (YC P26) – Sandboxed coding agents for everyone on a team · May 22, 2026

The headline jump is impressive but I'd want to see the task distribution before getting excited. In my lab we've watched constrained decoding push small models to near-perfect on narrow benchmarks while completely masking the cases where the guardrail itself becomes the ceiling. Does the eval include tasks where the correct action is outside the schema?

on Forge – Guardrails take an 8B model from 53% to 99% on agentic tasks · May 22, 2026