Project ideas from Hacker News discussions.

As A.I. makes law firms more efficient, clients ask: 'Where's my discount?'

📝 Discussion Summary (Click to expand)

We need to identify the 4 most prevalent themes in the opinions expressed in this Hacker News discussion. The discussion covers various topics: AI in healthcare (MRI diagnosis cost, upcoding, BCBS report), AI in legal industry (billing, privilege, disruption), AI in general (impact on software dev productivity, cost savings), and broader concerns about AI being used to increase costs, false positives, liability, and disruption of professions.

Let's parse the conversation:

  • Opening: Skyy93 about paying extra for AI diagnosis, concerns about false positives.
  • youniverse: imaging center charging $45 out of pocket.
  • pkaye: can you say no?
  • Skyy93: yes you can decline.
  • fluidcruft: questions about FDA-labeled AI diagnostics for MRI, says many are just workflow helpers.
  • elicash: BCBS report about hospitals using AI to find ways to charge more (upcoding).
  • hammock: suggests it's not bad for patients (increase standard of care).
  • judge2020: upcoding is prevalent, AI makes it easier, false secondary diagnoses.
  • franktankbank: UHG (UnitedHealth Group) happy with cost increases.
  • elicash: AI didn't change treatment (per BCBS).
  • judge2020: upcoding is old tactic, AI makes it easier; liability concerns if doctor dismisses AI suggestion.
  • elicash: from comment, suggests not running extra tests just charging.
  • dfxm12: healthcare already expensive, nickel and diming.
  • mcoliver: advice to get imaging data and run own analysis.
  • LastTrain: infrastructure spending.
  • ktallett: law companies investing in data centers.
  • LastTrain: if spend more on AI than payroll, clients won't get discount.
  • mullingitover: waiting for AI judgments, flash trials, flash litigation DDOS.
  • disantlor: YouTube link.
  • RachelF: Australia AI tax judgements (Robodebt scheme).
  • echoangle: says not really AI, just subtracting numbers.
  • plantain: Robodebt put onus of proof on taxpayer, no intelligence.
  • tehwebguy: Cyst O'Donavan joke.
  • edg5000: archive.is link.
  • throwaway89864: attorney-client privilege not applying to AI communications.
  • true_religion: they ruled it's like a journal/search engine.
  • judge2020: question about privilege, 5th amendment.
  • true_religion: depends on timing.
  • anon373839: chatting with Claude breaks confidentiality; self-hosted LLMs ok.
  • jryle70: email/phone considered confidential despite vendor inspection; expects precedent.
  • anon373839: weak analogy; AI companies monitor session data.
  • District5524: Latham & Watkins buying GPU clusters; not economical for smaller firms.
  • reply00r123: saved $20k, doesn't care.
  • tehwebguy: can't know saved until statute of limitations expire.
  • gedy: devs being 10x with AI, not getting raise.
  • byzantinegene: AI is 10x not devs, why should devs get raise?
  • besterman23: If AI reduces time for billable actions, reduces cost? Unless not reducing time.
  • judge2020: Reduces cost to provider, think it'll reduce patient bill?
  • besterman23: For lawyer yes, if task takes less time they bill less hours, cost less.
  • dabinat: Lawyers have templates, charge $1000 for 10 seconds work; fine with disruption.
  • hackyhacky: you'd rather pay $1000 to OpenAI for 5 ms vs human lawyer 10 sec.
  • kshmir: you probably pay cents in tokens.
  • hackyhacky: disruption business plan undercut then jack up prices (Uber, streaming).
  • dudisubekti: exploit opportunities, Darwinistic.
  • hackyhacky: disruptive businesses leveraging VC money to demolish established systems.
  • throwup238: if lawyer doesn't proofread, what's point? Scapegoat for AI?
  • hackyhacky: lawyers have ethical responsibility; prefer not money to Sam Altman consolidation.
  • somenameforme: concern not reasonable given many LLMs, open weight frontier models; valuations bubble; too big to fail logic.
  • atoav: you pay human because assume responsibility; if copy-paste without checking, no better than AI.
  • hackyhacky: comment not support AI instead of lawyers, but pointlessness of disrupting human lawyers.
  • tzs: if copy-paste without checking and AI wrong, can sue lawyer for malpractice.
  • knollimar: weird take on site with near zero marginal cost product.
  • dataflow: develop software for second client in 10 sec?
  • reverius42: each additional copy? under 2 sec if big.
  • warkdarrior: SaaS web app takes >10 sec to render?
  • Der_Einzige: lawyer jokes on point; they grift like tech workers pre-2022.
  • lurk2: not true at all.
  • georgeburdell: horrified at zero marginal cost legal work given strain on courts.
  • Razengan: courts could be replaced by AI too, with right to escalate to human judges.
  • shen: satisfied with chatbot vs talk to human?
  • Razengan: never, added prayer.
  • hibikir: law works because most people won't break it, disputes settle; if cost drops, need robo-judges or penalties.
  • somenameforme: self-inflicted by system expecting to plead not guilty when guilty; other countries treat guilty plea with remorse leniently.
  • nradov: not huge problem; people won't commit more crimes; civil side high-dollar disputes already litigated; growth in small claims scalable.
  • hermannj314: no evidence lawyers making job more efficient; burn industry, AI-only field, remove prejudice, misconduct, expense.
  • somenameforme: priesthood metaphor works; lawyers paid to decipher text they helped encipher.
  • nradov: legal profession hardly priesthood; contract language simple; people intimidated/lazy.
  • somenameforme: challenge to read bill HR6028; it's indecipherable, longer than Constitution.
  • tzs: looked at random sections, not undecipherable; comparison to Constitution length moot; Constitution is framework, bills implement details.
  • somenameforme: discretion: law lays out framework, interpreted with reasonable discretion; now laws written incomprehensible to non-lawyers.
  • tzs: need Federalist Papers etc to understand Constitution; bills are implementations complicated due to human complexity, interdependent economy, exploitation of ambiguity.
  • somenameforme: law uses broad strokes, relies on reasonable interpretation; obfuscation unnecessary.
  • JumpCrisscross: law is essentially a diff; USC easier to read.
  • alphasm: lawyers using AI to draft agreements, accurate on some matters, wrong on others due to lack of context; lawyers didn't review, causing post-signature issues, redo, inflated bills.
  • tsss: lawyers lobbying to outlaw AI-generated legal docs?
  • sillyfluke: they'd want to but would need to collude with AI companies; overcharging thing; public thinks they'd get discount with AI absurd; Oscars gift bags prenup template; moat is public illiteracy; but when stakes high, need lawyer who knows shit and accountable; find lawyers dependent on AI terrifying.
  • marcusb: outlaw AI-generated legal docs by non-lawyers.
  • fluidcruft: that happens even when AI not involved.
  • tiffanyh: any issues with tech industry selling software licenses? Lawyer selling contract template similar to software license/IP.
  • beaned: one transparent, other feels like lie.
  • atherton94027: still need to proofread doc and ensure applies.
  • thayne: akin to paying software contractor to make something specific, they give off-the-shelf product.
  • sa46: difference: lawyers charge by hour, software not based on creation time; if lawyer bills $1000 for 10 sec work smells fraudulent; if flat rate similar to software: purchasing IP, passing liability.
  • otterley: why fraudulent if fee agreement disclosed?
  • sa46: fee agreement hourly rate for producing legal doc; resulting doc looks like 10 sec; billed $1000 hours; either doc more complex (lawyer should have said) or lawyer inflated hours.
  • baq: they bill with hourly granularity.
  • crote: fine if getting hour of their time; if lawyer spent 1 min template, 10 sec fill, 20 min proofreading, nobody complaining; in reality they spend 1 min grabbing template, 10 sec fill, 3 min billing, 55 min double-dipping other clients.
  • knollimar: isn't it usually .1 hour increments?
  • otterley: we don't know fee agreement was hourly.
  • gspr: lawyer case time quantization inaccuracy; no smaller unit than hour; could aggregate smaller blocks.
  • lurk2: lawyers often don't proofread afterwards; hilarious seeing complaints about lawyers overpaid/incompetent on forum dedicated to industry unregulated until a month ago; lawyer can be disbarred for mistakes; can you?
  • siva7: did i miss something? are developers regulated since last month?
  • baq: no, they irrelevant.
  • fcarraldo: pretty sure more Wild West; engineers not reading code.
  • mrheosuper: not fair comparison; template like source code takes time/money to write/maintain.
  • MichaelRo: salaryman lament against contract workers (plumbers, car mechanics, lawyers): he makes $25/hour, why don't they charge similar or less? Because unlike steady stream, they only make a living if called.
  • otterley: plumber analogy: tapping pipe $5, knowing where to tap $495.
  • revolvingthrow: plumber didn't know where to tap or didn't care; can tap random spot for <$5.
  • rpunkfu: they can fill random doc; expertise which template to use fills scenario; subtle nuances you don't want to do yourself.
  • analog31: other version: customer $500 for 10 min work? > I make as brain surgeon. Plumber: more than I made as brain surgeon too.
  • harpersealtako: pipe tapping $5, knowing where to tap $95, taking liability $400; ChatGPT can tell where to tap but can't sue OpenAI if wrong.
  • glenngillen: lawyer admitted they'd not defend in court; if issues, you send to them and they charge rate; need clause to protect themselves, send stern emails; if ends up in court, need litigation specialist.
  • fbrncci: mostly a thing in West; in Asia lawyers email back/forth weeks, get right docs/advice, charge when actual work needed; shocking because would have spent $1000+ in EU/US.
  • xgbi: metaphor translates to software: fix problem for $10k coding time, then sell 1000s clients same software for $99, 0s work.
  • palmotea: lawyer templates like library/SaaS; if you/employer sells software, you doing same thing.
  • tpoacher: except software no expectation client debugs before accepting; bugs after purchase treated as client's fault.
  • fcarraldo: worse; software users randomly flagged as testers without knowledge/consent; they get changes and pay for privilege; have to contact support when breaks.
  • palmotea: client does need to test; B2B licenses unknown; B2C user responsible for output, vendor liability capped (refund purchase price); if Excel bug costs money, user responsible, Microsoft not paying.
  • Havoc: depends on lawyer/work; if just copy-paste templates then never needed lawyer; paralegal at best.
  • georgefrowny: UK conveyancing solicitors: 90% shuffle standard paperwork, chase other solicitors; could be replaced by £10 AI + couple hours human time.
  • plastic-enjoyer: wonder how easily lawyers let industry disrupted; think lawyers have means/edge to push back.
  • fluidcruft: now AI will proofread and quality improves.
  • Animats: law firms rolled over by AI; inefficient firms replaced like Expedia did to travel agents.
  • nomilk: case for many industries; software slowly ate property managers work; remote inspection, automated rent payments; but number of property managers didn't decrease, price landlords paid didn't decrease.
  • apical_dendrite: luxury buildings use AI agents to schedule tours, answer questions; tours self-serve; crappy experience; had to call AI agent to figure out how to get in front door, failed; saw groups wandering unescorted; not safe feeling.
  • neya: AI is just a tool like powered screwdriver; are you also going to ask me for discount because I used powered screwdriver?
  • mpalmer: yes, because I'm expecting you to compete.
  • tahoeskibum: exactly, somebody using power tools can do more in an hour than manual tools; that's real discount.
  • neya: doesn't matter, people pay for output quality, not tools used; that's why pay almost $2000 for iPhone that costs $200 and <2h to make.
  • mpalmer: category error comparing biggest tech firms to all law firms; $200 to make $2000 phone flat-out wrong; iPhone 16 cost pushing $600, sold for no more than $1500; when obvious to consumer you save massive time/money using new tech and keep margin, choice is easy: go to competitor who wants business more than keeping savings.
  • neya: using powered screwdriver somehow incompetent? flawed logic.
  • DrSAR: conversely, I'll pay you more if you grab a rake and not a leaf blower.
  • preommr: since posting pointless gotchas: do we pay more to lawyer if they handwrite doc using expert penmanship?
  • byzantinegene: yes, because billing based on hours.
  • neya: no, I'm billing you for output of my work, not hours.
  • lizknope: know almost nothing about law firms; but for other industry if 1000 competitors making similar products then compete on price; reduce price get more customers; if want new PC would rather have IBM making PCs setting price or thousand clone companies competing; so some new law firm may want to disrupt industry with new model.
  • mikert89: spoke with banker at well-known investment bank NYC; they told law firm want fees cut in half or walk; to top 5 law firm; lawyers trying to charge same prices when both sides know working half hours; even at premium top end where engagement bill can run up to 30 million.
  • lurk2: lawyers trying to charge same prices when both sides know working half hours – absurd claim without evidence.
  • mathisfun123: 80% of HN comments.
  • anal_reactor: surprisingly fantastic result considering literally all other websites; ngl guilty myself too.
  • mrheosuper: absurd claim presented without evidence.
  • mathisfun123: your epistemic barometer way off if you think claiming "lawyers are now doing 50% less work" same as "people on internet make shit up all the time" lolol.
  • jareklupinski: source: my lawyer friend now orders two martinis for lunch instead of one.
  • mikert89: you want me to post their name?
  • throwyawayyyy: aren't bankers charging same prices they were before computerization?
  • mikert89: value from bankers comes from relationships, sales, weight in doing good IPO; not about hours worked.
  • Henchman21: are we not still paying to “build out” payment networks for Visa and Mastercard?
  • cupofnotjoe: even if had evidence, well-known investment bank never using top 5 law firm because of rates charged.
  • sensanaty: because bankers famously aren't greedy pigs, and they definitely lowered prices for customers after efficiencies with computerization, right?
  • edoceo: one thing you get from legal council is they'll be ready to defend contracts/agreements they've made for you; might be hard to find defense/offense on AI slop contracts; seen how AI does code, imagine inconsistency in legal docs.
  • narmiouh: but ones drafting docs aren't ones fighting for contract in court; one who drafted sloppy is happily on to next unsuspecting client, while you bear cost of fighting with and paying for brand new lawyer in courts; no good processes to sue lawyers for malpractice like doctors; negligence can cause entire companies to go down, perpetrators almost never pay price.
  • socializer: outside court, significant proportion of what law offices do already extremely streamlined; stuff like wills, deeds, incorporation paperwork all template-based, remaining work by paralegals then billed at outrageous rates; not sure if AI makes whole lot cheaper; maybe makes harder to pretend; but there's good reason lawyers can keep charging fees: because they have a say on who can practice profession; unlike software engineers and like doctors, they're in charge and will not allow autonomous systems to represent you or otherwise "practice law"; that'd be suicide; also, for many services downward price pressure already exists to some extent, low-stakes virtual legal firms licensed in multiple states provide forms you can fill and file for $50 a pop instead of $500 a pop; but most consumers still prefer local law office; not sure will change with AI; know quite a few people keen to self-diagnose using Google or ChatGPT, but not as many inclined to vibecode articles of incorporation.
  • classified: > That'd be suicide. And yet many software developers are hyping themselves into obsolescence by pretending you can't write code by hand anymore.
  • Razengan: lawyers and politicians are first things that ought to be replaced by AI, honestly.
  • classified: Remember that it can always get worse.
  • Razengan: we've watched Terminator and The Matrix. We're ready.
  • springtimesun: partner spent many years in big law and now sits across table on client side; billable hour problematic longer than AI; but many forget (including companies paying bill) that

🚀 Project Ideas

Patient AI Diagnosis Transparency & Second Opinion Portal

Summary

  • Lets patients view AI‑generated diagnostic suggestions from their imaging reports, including confidence scores and alternative interpretations, and request an independent second‑opinion read.
  • Gives patients the information and leverage to question unnecessary AI add‑on charges, reducing surprise costs and building trust in AI‑assisted care.

Details

Key Value
Target Audience Patients undergoing MRI/CT scans who are offered AI‑diagnosis add‑ons
Core Feature Upload radiology report (or pull via FHIR) → see AI suggestions, confidence, alternatives → request a second read from a board‑certified radiologist
Tech Stack React frontend, Node.js/Express backend, FHIR‑compatible API, Python for confidence scoring, optional open‑source rad‑AI model for second opinion
Difficulty Medium (HIPAA compliance, integration with hospital systems)
Monetization Revenue-ready: per‑report fee ($9‑$19) or subscription for frequent imaging

Notes

  • Addresses Skyy93’s frustration: “getting an AI diagnosis cost extra… can understand why people would be hesitant to pay extra when they hear about false positives.”
  • Responds to fluidcruft’s concern: “Are there even any FDA labeled AI diagnostics for MRI? … the things I've seen gaining traction are very specific measurements….” by providing transparency and a path to verify AI output.
  • Empowers patients to avoid being nickel‑and‑dimed for unreliable AI, a point raised by dfxm12.

Self‑Hosted LLM Legal Privilege Suite

Summary

  • A turnkey Dockerized kit that lets law firms run open‑weight LLMs entirely on‑premise or in a private VPC, guaranteeing no client data leaves the firm’s control.
  • Includes audit logging, role‑based access, and seamless plug‑in to existing document‑assembly tools so lawyers can safely use AI for drafting, research, and proofreading without breaking attorney‑client privilege.

Details

Key Value
Target Audience Small‑to‑mid law firms and solo practitioners worried about privilege when using consumer AI services
Core Feature One‑click deployment of Llama‑3 / Mistral models with encrypted storage, prompt logging, and a chat‑based UI for legal research and drafting
Tech Stack Docker (or Kubernetes), Python FastAPI, HuggingFace Transformers, PostgreSQL for audit logs, optional NVIDIA GPU support, React admin UI
Difficulty High (requires GPU driver setup, model weight management, security hardening)
Monetization Revenue-ready: annual license per seat ($1,200) + optional support contract

Notes

  • Hits anon373839’s warning: “Chatting with Claude breaks confidentiality… self‑hosted LLMs without breaking privilege.”
  • Matches District5524’s observation that “for anyone smaller than them it is worth spending on HW… geofenced datacenters etc.” by offering an affordable, self‑hosted alternative.
  • Gives firms a concrete way to retain control of AI while still gaining productivity gains, a need repeatedly mentioned in the thread.

AI Billing Upcode Detector

Summary

  • Scans hospital bills and extracts to flag secondary diagnoses that appear only after AI‑driven chart review and lack supporting clinical evidence, helping patients and insurers spot potential upcoding.
  • Provides plain‑English explanations of why each flagged charge may be unjustified and suggests next steps (request itemization, file dispute).

Details

Key Value
Target Audience Patients receiving high‑cost inpatient bills, medical bill auditors, self‑insured employers
Core Feature Upload PDF/XML bill → tool cross‑references diagnosis codes with note timestamps, labs, and imaging to detect AI‑added secondary diagnoses lacking corroboration → outputs a risk score and plain‑language rationale
Tech Stack Python backend, medspaCy/NLP for clinical text, FHIR parser for EHR data, React UI, optional integration with hospital billing APIs
Difficulty Medium (needs medical coding knowledge but can start with rule‑based detection)
Monetization Revenue-ready: per‑bill audit fee ($25) or SaaS subscription for employers ($199/mo)

Notes

  • Directly tackles elicash’s BCBS finding: “hospital systems are increasingly billing… secondary diagnoses… driving nearly $1 billion in additional healthcare costs… more than 60%… using AI‑enabled technologies.”
  • Addresses judge2020’s worry about “false secondary diagnoses… makes hospitals… a boatload of money” and the difficulty of contesting AI findings.
  • Responds to dfxm12’s complaint about feeling “nickel‑and‑dimed” by giving patients a concrete way to audit and challenge suspect AI‑inflated charges.

Smart Legal Template Generator with Explainability

Summary

  • Guides non‑lawyers through a questionnaire to produce customized legal documents (NDAs, service agreements, incorporation paperwork) using vetted templates while highlighting which clauses are boilerplate and which need attention.
  • Offers an optional “lawyer‑in‑the‑loop” review where a licensed attorney checks the document for a fixed low fee, giving users confidence without the typical $1000 markup for simple edits.

Details

Key Value
Target Audience Entrepreneurs, freelancers, and small businesses needing routine legal docs but wary of overpriced lawyer fees
Core Feature Interactive form → dynamic template filling → real‑time plain‑language explanation of each clause → one‑click purchase of attorney review
Tech Stack React frontend, Node.js/Express backend, template engine (Handlebars/Jinja), clause library database, Stripe for payments; optional lawyer‑marketplace integration
Difficulty Low‑Medium (mostly UI/template work; lawyer network can be sourced via existing platforms)
Monetization Revenue-ready: $20 per document generation + $50 for lawyer review, or $15/mo subscription for unlimited docs

Notes

  • Mirrors palmotea’s experience: “My experience with lawyers is that they often have templates… take 10 seconds to fill… then charge you $1000… I’m totally fine with that industry getting disrupted.”
  • Supports davidsubekti’s view that “if you need a legal document… you can take that and run it through your own analysis…” by giving users a guided, transparent way to create and vet documents.
  • Provides a low‑cost, trustworthy alternative to the “$1000 for 10 seconds of work” model frequently criticized in the thread.

Read Later