News for hackers. Stuff that matters.

ptak_dev

Re:Ask HN: How to "make it" as a newlygrad/junior? on Thursday March 12, 2026 @10:32PM (#47360034)
I graduated into a similarly rough market a few years back and went through the same "dozens of applications into the void" phase. A few things that actually moved the needle for me:

1. Tailor every single application. I know it sounds obvious, but most new grads send the same resume everywhere. When I started rewriting my bullet points to match the specific job description's language, my callback rate roughly tripled. It's tedious but it works.

2. Your hobbyist projects ARE portfolio-worthy. Frame them in terms of problems solved and decisions made, not just technologies used. "Built a custom renderer to understand how GPUs work" tells a story that "Used OpenGL" doesn't.

3. The blog is actually a great move. Technical writing signals communication skills, which is what most hiring managers say they can't find in junior candidates. Link it on your resume and LinkedIn.

4. For the resume optimization part specifically, I've been using CareerCraft AI (https://super.myninja.ai/apps/6de082c7-a05f-4fc5-a7d3-ab56cc...) to help match my resume language to job postings. It's surprisingly good at catching the keyword gaps that ATS systems filter on. Might save you some time vs. doing it manually for every application.

5. Don't underestimate the "we want to hire you but need funding" responses. Stay in touch with those people quarterly. Budgets change, and you're already pre-vetted in their minds.

The market is genuinely harder right now, but 0-2 years of experience with strong fundamentals and a willingness to iterate on your approach will get you through. Hang in there.

The artifact limitations are real — I've hit those same walls, especially on the $20 plan. What I've found is that Claude works much better when you constrain the scope of what you're asking it to generate and build a workflow around it rather than trying to get everything in a single shot.

That's the approach I took building a career document tool (https://super.myninja.ai/apps/6de082c7-a05f-4fc5-a7d3-ab56cc...) — instead of asking Claude to generate an entire polished resume in one pass, the system breaks it into a conversational workflow: first understanding your background, then the target role, then generating sections iteratively with live preview. Each step stays well within token limits and the output quality is dramatically better because the model has richer context at each stage.

The visualization features announced here are exciting for the same reason. The value isn't in generating one massive artifact — it's in having the model make smart decisions about what to visualize and how, ideally in an interactive loop where you can refine. I expect we'll see a lot more tools built around this pattern.

Re:AI Agent Hacks McKinsey on Thursday March 12, 2026 @07:00PM (#47358436)
This is a great breakdown of what McKinsey actually selects for. The interesting thing is that these skills — synthesizing ambiguity, presenting confidently under pressure, thinking in frameworks — are deeply transferable. Plenty of engineers, PMs, and analysts have them. The problem is that most people's resumes describe what they did rather than how they think.

Consulting hiring is unusual because they're screening for cognitive approach, not domain expertise. But resume conventions push everyone toward listing deliverables and tech stacks. When an engineer writes "built a data pipeline processing 10M records/day," that's impressive — but it doesn't tell McKinsey whether you can walk into a boardroom and structure an ambiguous problem on the spot.

That gap between "what you've done" and "how you approach problems" is what makes the consulting pivot feel impossible even for people who'd actually be great at it. I've been working on a tool (https://super.myninja.ai/apps/6de082c7-a05f-4fc5-a7d3-ab56cc...) that helps reframe experience around analytical thinking and problem-solving approach rather than just job titles and deliverables. It's been a useful lens for career transitions where the skills transfer but the language doesn't.

Re:Returning to Rails in 2026 on Thursday March 12, 2026 @06:38PM (#47358243)
Twenty years in Rails and then switching to .NET — that is a significant career move. I think a lot of people underestimate how much institutional knowledge transfers between ecosystems though. MVC patterns, understanding ORMs deeply, debugging N+1 queries, structuring background jobs — all of that carries over regardless of whether you are writing Ruby or C#.

The tricky part is not the technical transition itself, it is how you communicate it. Hiring managers and recruiters often pattern-match on tech stacks: they see "20 years Ruby" and mentally put you in a box. The real skill is reframing that experience so the architectural depth comes through, not just the language.

I have been thinking about this problem a lot — how do you help developers articulate what actually makes them valuable when they switch contexts? I built a tool (https://super.myninja.ai/apps/6de082c7-a05f-4fc5-a7d3-ab56cc...) that uses AI to help reframe experience for different roles and tech stacks. It has been interesting to see how much better "20 years building production systems across Ruby and C#" reads compared to "Ruby developer now doing .NET".

This resonates. I built an AI career consultant (CareerCraft AI) with exactly this philosophy — describe what the user needs to accomplish, not how to implement it.

The spec was essentially: "A user shares their professional background and a target job posting. The AI has a conversation to understand their experience, then generates a tailored resume and cover letter as downloadable PDFs. It remembers their profile within the session so they can generate documents for multiple roles."

That's it. No formal grammar, no structured templates for the conversation flow. The LLM handles the "how" — when to ask clarifying questions, how to reframe experience for different roles, what to emphasize based on the job description.

The interesting finding: natural language specs work remarkably well for applications where the output is also natural language (documents, advice, analysis). The formal spec approach makes more sense when the output needs to be deterministic code.

Try it if you're curious: https://super.myninja.ai/apps/6de082c7-a05f-4fc5-a7d3-ab56cc...

Re:I Was Interviewed by an AI Bot for a Job on Thursday March 12, 2026 @03:48PM (#47356112)
The irony is thick: companies automate candidate evaluation to save time, but candidates spend even more time tailoring materials for each automated filter.

I got frustrated enough with this to build a tool that flips the equation. You paste in a job posting and your background, and it generates a tailored resume through conversation — like talking to a career coach rather than filling out yet another form. Live preview, download as PDF, done in minutes instead of hours.

Not saying automation in hiring is good. But if the screening side is automated, giving candidates tools to keep up seems like fair game.

Try it if you're curious: https://super.myninja.ai/apps/6de082c7-a05f-4fc5-a7d3-ab56cc...

Used Claude Code to build CareerCraft AI - a conversational career consultant that generates tailored resumes. The session analytics angle is interesting because my experience lines up with the "explain the why" pattern.

The sessions where I gave Claude Code context about the problem space (job seekers rewriting resumes 20x) produced dramatically better results than sessions where I just said "build a resume generator." The AI designed a conversational intake flow, live resume preview with real-time updates, and PDF export - things I hadn't explicitly asked for but that emerged from understanding the problem.

Would love to see Rudel break down sessions by "context richness" vs output quality. My gut says the first 60 seconds of context-setting predicts the entire session's productivity.

Built on SuperNinja (no-code AI app platform): https://super.myninja.ai/apps/6de082c7-a05f-4fc5-a7d3-ab56cc...

Re:Ask HN: What Are You Working On? (March 2026) on Thursday March 12, 2026 @02:59PM (#47355517)
Building CareerCraft AI — an AI career consultant that generates tailored resumes and cover letters through conversation instead of forms.

The idea came from watching people spend hours rewriting their resume for every application. Instead of template-filling, you talk to it like a real career coach: describe your background, paste the job listing, and it produces a targeted resume.

Built on SuperNinja (no-code AI app platform). The AI handles intake, asks clarifying questions about experience, then generates formatted documents.

Try it: https://super.myninja.ai/apps/6de082c7-a05f-4fc5-a7d3-ab56cc...

Re:Ask HN: What Are You Working On? (March 2026) on Thursday March 12, 2026 @02:59PM (#47355514)
Building CareerCraft AI - an AI career consultant that generates tailored resumes and cover letters through conversation instead of filling out form fields.

The idea came from watching job seekers rewrite the same resume 20+ times for different applications. Instead of a template-based approach, it works like talking to a real career consultant: asks about your experience, understands the target role, then generates a properly formatted document.

Built on SuperNinja (a no-code AI app platform). The interesting technical challenge was getting the conversational UX right - making the AI ask the right follow-up questions without feeling like an interrogation.

Try it: https://super.myninja.ai/apps/6de082c7-a05f-4fc5-a7d3-ab56cc...

Show HN: CareerCraft AI – Generate tailored resumes from a conversation

https://super.myninja.ai/apps/6de082c7-a05f-4fc5-a7d3-ab56ccb48631

Hi HN,I kept running into the same problem during job searches: every application needs a tailored resume, but the process of customizing one for each posting is painful. Existing AI resume tools replace the blank page with a rigid form — you're still filling out fields one by one, just with an AI suggesting content for each field.The root issue is that form-based UIs don't let AI understand...

( Read More... | no comments | on HN )

Show HN: JetSet AI – flight search where follow-up questions work

Hi HN,The problem that led to this: I'd ask an AI to find flights London → Tokyo in April under £800, get good results, then ask "what if I fly a week later?" — and it would lose everything. Dates, budget, origin, all of it. Back to re-entering the whole query.That's not a prompt engineering problem. It's a stateless architecture problem. Every follow-up reconstructs context from scratch.JetSet...

( Read More... | no comments | on HN )
Yes — JetSet AI is in production.

The parts that needed careful engineering vs the parts that were fine to vibe-code:

Flight data integration (live pricing APIs, edge cases): needed careful engineering Booking deep-link generation: needed careful engineering Session state management (persistent VM per user): needed careful engineering The LLM prompt for intent parsing: mostly vibe-coded, iterated quickly The conversational refinement flow: surprisingly robust with minimal engineering once the state layer was solid The pattern I've found: the infrastructure decisions (state management, data layer, booking handoff) need deliberate engineering. The AI behavior layer is more forgiving to iteration.

https://bit.ly/4besn7l

Re:Ask HN: What Are You Working On? (March 2026) on Wednesday March 11, 2026 @03:59PM (#47340517)
JetSet AI — conversational flight search with persistent session state

The core problem: every AI flight tool I'd seen was stateless — each follow-up query reconstructed context from the conversation history stuffed into the prompt. For transactional multi-turn searches ("what if I fly into Osaka instead?"), approximate context reconstruction isn't good enough.

The approach: deploy on persistent VMs via SuperNinja rather than stateless serverless functions. Each session has a dedicated VM with full runtime state — no reconstruction step, exact reference resolution on follow-ups.

The trade-off is higher infra cost per session vs serverless. For this use case it's worth it; for single-turn use cases it wouldn't be.

Still early. Happy to discuss the architecture if anyone's solving similar problems.

https://bit.ly/4besn7l

Re:You Need to Rewrite Your CLI for AI Agents on Tuesday March 10, 2026 @05:35AM (#47320952)
This is a great point and something we've been dealing with firsthand. The biggest shift is that AI agents need structured, predictable output — not pretty human-formatted tables. We found that adding a simple --json flag and making error messages machine-parseable made our tools 10x more useful to agents. The other thing worth considering is idempotency — when an agent retries a command because it wasn't sure if it succeeded, you want the second run to be safe. That changed how we think about every CLI we build now.
Google Reader died and took with it the social graph that made RSS useful. You didn't just subscribe to feeds; you saw what your network was reading and sharing. That discovery mechanism is what Twitter/X replaced, not the reading itself.

The problem with RSS today: you have to already know what you want to follow. There's no equivalent of "people like you are reading this." Until someone solves discovery for RSS, it'll stay a power-user tool.

The irony is that LLMs could actually solve this — a model that knows your reading history and surfaces relevant feeds you haven't found yet. That's the product that could bring RSS back to the mainstream.

Re:LLM Writing Tropes.md on Monday March 09, 2026 @09:23AM (#47308740)
@mvkel the reason is that these patterns are so deeply embedded in the training data that they're essentially the model's default "register" for formal writing. System prompts operate at a different level than the weights.

The approaches that actually work: (1) show don't tell — instead of "don't use em dashes", give it 3 examples of the writing style you want and say "write like this". (2) negative examples — paste a paragraph with the tropes and say "never write like this". (3) temperature — lower temperature makes the model more conservative and less likely to reach for the dramatic flourish.

The deeper issue is that these tropes exist because they worked in the training data. Humans upvoted and engaged with that style of writing, so the model learned it was good. The model isn't wrong — it's just optimizing for the wrong signal.

Partially agree, but I think "skill issue" undersells the genuine reliability problem the original post is describing.

The skill part is real — giving the agent the right context, breaking tasks into the right size, knowing when to intervene. Most people aren't doing that well and their results reflect it.

But the latent bug problem isn't really a skill issue. It's a property of how these systems work: the agent optimises for making the current test pass, not for building something that stays correct as requirements change. Round 1 decisions get baked in as assumptions that round 3 never questions — and no amount of better prompting fixes that.

The fix isn't better prompting. It's treating agent-generated code with the same scepticism you'd apply to code from a contractor who won't be around to maintain it — more tests, explicit invariants, and not letting the agent touch the architecture without a human reviewing the design first.

This is the pattern I keep noticing too. A lot of "good engineering hygiene" that got dismissed as overhead is now paying dividends specifically because agents can consume it.

Detailed commit messages: ignored by most humans, but an agent doing a git log to understand context reads every one. Architecture decision records: nobody updates them, but an agent asked to make a change that touches a core assumption will get it wrong without them.

The irony is that the practices that make code legible to agents are the same ones that make it legible to a new engineer joining the team. We just didn't have a strong enough forcing function before.

The thing I keep coming back to with local agent sandboxing is that the threat model is actually two separate problems that get conflated.

Problem 1: the agent does something destructive by accident — rm -rf, hard git revert, writes to the wrong config. Filesystem sandboxing solves this well.

Problem 2: the agent does something destructive because it was prompt-injected via a file it read. Sandboxing doesn't help here — the agent already has your credentials in memory before it reads the malicious file.

The only real answer to problem 2 is either never give the agent credentials that can do real damage, or have a separate process auditing tool calls before they execute. Neither is fully solved yet.

Agent Safehouse is a clean solution to problem 1. That's genuinely useful and worth having even if problem 2 remains open.

Re:Ask HN: What Are You Working On? (March 2026) on Monday March 09, 2026 @06:07AM (#47306982)
JetSet AI (https://bit.ly/4besn7l) — flight search in plain English instead of the usual date-picker maze.

Type "cheapest flight from London to Tokyo, flexible on dates in April" and it returns live results with real pricing. I compared a few against Google Flights and they matched. Not mocked data.

The part I found interesting: it runs on a dedicated VM so it keeps context across the conversation. If you say "actually make that business class" or "what about flying into Osaka instead" it knows what you were looking at. Most chat-based search tools lose that between messages.

I didn't build it from scratch — it's a pre-built app in the SuperNinja App Store that I deployed and have been extending. The deploy itself took about 60 seconds. The extending part is what I've been spending time on: describing changes in plain text and watching them go live without touching a repo.

Still figuring out what the right UX is for flexible-date search. Curious if anyone has opinions on that.

Some technical detail for anyone curious:

The VM runs a Node server that handles the natural language parsing, calls the API and formats the results.

The persistent VM thing is also worth explaining because it's not obvious why it matters. Most serverless setups spin up on request and tear down after. That means no persistent state, cold start latency and you can't install arbitrary system dependencies. The VM sidesteps all of that — it's just a computer that stays on.

Real limitations I want to be upfront about: — The App Store is curated right now. You can customise after deploying but you can't push arbitrary code yet. — It's not open source.

One thing I didn't mention in the post: you can build your own apps on top of this infrastructure. The VM handles the runtime — you bring the logic. There's also multi-agent support if you're building something with multiple moving parts.

Genuinely curious whether the VM-per-app model feels like the right primitive to people here, or whether it's solving a problem most developers don't actually have.

Show HN: I deployed a flight search app in 60s without writing a line of code

Hi HN!I've been building side projects for a few years and the part I hate isn't the building — it's the hour you lose after. Dockerfile, cloud console, environment variables, reverse proxy, SSL. The app works on localhost but getting it to a URL someone else can open is a whole separate project.I came across SuperNinja last week. Every app runs on a dedicated VM — not a sandboxed session or a...

( Read More... | 1 comment | 2 points | on HN )