Hacker News

Hi HN — I’m Prateek Rao. My cofounders and I built Cortexa, which we describe as a Bloomberg terminal for agentic memory.

A pattern I keep seeing: when agents misbehave, most teams iterate on prompts and then “fix” it by plugging in a memory layer (vector DB + RAG). That helps sometimes — but it doesn’t guarantee correctness. In practice it often introduces a new failure mode: the agent retrieves something dubious, writes it back to memory as if it’s truth, and that mistake becomes sticky. Over time you get memory pollution, circular hallucination loops, and debugging turns into log archaeology.

What Cortexa does:

1. Agent decision forensics (end-to-end “why”): trace outputs/actions back to the exact retrievals, memory writes, and tool calls that caused them.

2. Memory write governance: intercept and score memory writes (0–1), and optionally block/quarantine ungrounded entries before they poison future runs.

3. Memory hygiene + vector store noise control: automatically detect and remove near-duplicate / low-signal entries so retrieval stays high-quality and storage + inference costs don’t creep up.

Why this matters: Observability is the missing layer for agentic AI. Without it, autonomy is fragile: small errors silently compound, deployments become risky, and engineering cost goes up because failures aren’t reproducible or attributable.

Who this is for: 1. Teams shipping agentic workflows in production 2. Anyone fighting “unknown why” failures, memory pollution, or runaway context costs 3. Engineers who want auditability + faster debugging loops

Site: https://cortexa.ink/

Would love feedback from anyone running agents at scale: 1.What’s the most painful agent failure mode you’ve seen in production? 2.What signals would you want in an “agent terminal” (retrieval diffs, memory blame, tool-call traces, alerts, etc.)?

Comments URL: https://news.ycombinator.com/item?id=47228173

Points: 6

# Comments: 1

These days I go to new/ and it is overwhelming. Large quantity of low quality submissions.

Proposed solution: limit submissions to accounts with >X karma (eg. 500 karma) OR older than Y years (eg 5 years).

Both of these scale into the future as one can reach both conditions without ever submitting (eventually).

Comments URL: https://news.ycombinator.com/item?id=47228169

Points: 2

# Comments: 2

Article URL: https://www.ntik.me/posts/small-teams

Comments URL: https://news.ycombinator.com/item?id=47228159

Points: 3

# Comments: 0

Article URL: https://afterlive.ai

Comments URL: https://news.ycombinator.com/item?id=47228137

Points: 2

# Comments: 0

Article URL: https://deepexplain.dev/dirac-notation/

Comments URL: https://news.ycombinator.com/item?id=47228131

Points: 2

# Comments: 0

Article URL: https://www.nature.com/articles/s41562-026-02423-6

Comments URL: https://news.ycombinator.com/item?id=47228128

Points: 2

# Comments: 2

Article URL: https://www.wsj.com/livecoverage/iran-israel-us-strikes-2026/card/u-s-marines-fire-on-protesters-in-karachi-pt1L3kaX1WsKliYRT66M

Comments URL: https://news.ycombinator.com/item?id=47227953

Points: 1

# Comments: 0

Article URL: https://www.biorxiv.org/content/10.64898/2026.02.05.703998v1

Comments URL: https://news.ycombinator.com/item?id=47227948

Points: 1

# Comments: 0

Article URL: https://www.praetorian.com/blog/introducing-augustus-open-source-llm-prompt-injection/

Comments URL: https://news.ycombinator.com/item?id=47227919

Points: 1

# Comments: 1

Article URL: https://twitter.com/eoghan/status/2028522852044206258

Comments URL: https://news.ycombinator.com/item?id=47227913

Points: 1

# Comments: 0

Starcraft2 is an old game, but it's always lacked a way to visualize game replays outside of the game itself.

I built a replay rendering engine from scratch using the replay files and Claude Code.

The replay files contain sampled position coordinates and commands that the player inputs. So I built an isometric view using the map and overlayed unit icons over the map, then interpolated the positions that units move in over time.

I also extracted additional metrics from the game data as well - some are derived on top of other metrics.

Finally, I pass all this context into a LLM for it to critique gameplay and offer strengths and improvements per player.

It's not perfect, but a good starting point to iterate and improve

Let me know what you think!

Comments URL: https://news.ycombinator.com/item?id=47227902

Points: 1

# Comments: 0

Article URL: https://www.salon.com/1998/05/04/feature_325/

Comments URL: https://news.ycombinator.com/item?id=47227892

Points: 1

# Comments: 0

Article URL: https://catalinionescu.dev/ai-agent/building-my-own-canva-over-the-weekend/

Comments URL: https://news.ycombinator.com/item?id=47227891

Points: 1

# Comments: 0

Article URL: https://sites.socsci.uci.edu/~ddhoff/interface.pdf

Comments URL: https://news.ycombinator.com/item?id=47227847

Points: 1

# Comments: 0

The Support Agent Who Never Burns Out Human-like AI teammates are quietly solving the problem that broke customer service. Meet Sarah. Sarah is your best customer support agent. She knows your product cold, handles difficult customers with patience, and resolves tickets faster than anyone on the team. She also called in sick Monday, runs on fumes by Thursday, and quit last April right after you finished training her replacement. This is the story nobody tells about customer service. The quiet structural collapse underneath the chatbot failures and the CSAT scores. The Math Has Never Worked Call center turnover runs 30 to 45% annually, more than double any other industry. Replacing one agent costs $10,000 to $20,000. Across a 100-person team, that's over $1M in churn before you've served anyone well. • 87% of contact center workers report high stress on the job • 59% are at active risk of burnout • 77% say workload has increased compared to the previous year

US businesses risk losing $856 billion annually to poor customer service, not because companies don't care, but because the system is structurally broken. What Happens at 2 AM Your customers don't keep business hours: • A prospect in a different time zone has a pre-sale question • A customer spots a billing error on Sunday evening • A new user is stuck in onboarding at midnight, about to close the tab

78% of customers have abandoned a purchase due to poor service. The problem isn't AI. It's the wrong kind — built for deflection, not resolution. What's Actually Changed A new category has emerged: conversational video AI teammates. Not animated chatbots with a moving mouth. Digital humans capable of: • Holding full conversations with dynamic facial expressions and natural eye contact • Responding with emotional tone that adapts to the customer in real time • Pulling contextually aware answers from your actual company data

Leading platforms now deliver under 1 second end-to-end latency, under 80ms speech-to-avatar response, and unlimited concurrent sessions. Sarah and Maya: Side by Side It's 11:47 PM. Maya notices a duplicate charge and visits your support page. Old model: • Chatbot asks clarifying questions, fails to pull account data • Offers a help article, tells her to call back during business hours • Maya disputes the charge through her bank. You lose the relationship.

AI teammate model: • Digital human appears immediately, greets Maya by name • Pulls account info in real time, confirms the duplicate charge • Issues the refund and sends confirmation, all within the same session

Total time: under four minutes. First-contact resolution. No ticket. No human agent needed at midnight. AI teammates don't replace Sarah. They protect her from the 80% of tickets that were burning her out. What the Numbers Say • AI-assisted support improves issues resolved per hour by 14% • Average return of $3.50 for every $1 invested, top performers hitting 8x • Gartner projects $80 billion in global call center labor cost reductions • AI customer service market growing from $12B (2024) to $47.82B (2030)

64% of consumers say they're more likely to trust AI customer service if it exhibits human-like traits. That's the trust gap text chatbots have never closed. Platforms like Trugen AI are building exactly this, where AI teammates see, hear, and respond with genuine presence. Sarah Gets to Stay Real-time AI teammates absorb the volume: • Billing disputes, order status checks, password resets • Product FAQs and after-hours inquiries

Sarah handles what genuinely needs her: • Escalations, complex problems, high-value accounts • Customers in real distress who need a human

Her job improves. Her burnout risk drops. She stays. If you're exploring what this looks like for your team, Trugen AI is worth a look as a starting point. Tags: Customer Service, AI Agents, Digital Humans, Conversational AI

Comments URL: https://news.ycombinator.com/item?id=47227826

Points: 1

# Comments: 0

Article URL: https://carnegieendowment.org/emissary/2026/03/iran-china-us-intervention-strategy

Comments URL: https://news.ycombinator.com/item?id=47227824

Points: 2

# Comments: 0

I created a personal AI gateway that runs on your local machine to provide guardrails for any service that communicates with popular LLMs. You can run this as a proxy for openclaw to inject prompts, filter PII, warn if there are jailbreak attempts. Can also work in a distributed fashion.

It creates a personal access token (PAT) based on your API keys from Anthropic, OpenAI or others, then injects policies. It does not store secrets, but instead stores a reference to environment variables storing those values.

Another benefit - if you execute Claude code or Codex using the run command, you can also record session token usage and the prompt traffic. This allows you to store it with your github project as a shared memory.

Run the binary using 'run' for a context that lasts for the command, or use 'start/stop' to initiate the proxy at localhost:8443. There is a web UI you can access to gather stats as well.

Comments URL: https://news.ycombinator.com/item?id=47227802

Points: 1

# Comments: 0

Article URL: https://www.crowdsupply.com/baochip/dabao

Comments URL: https://news.ycombinator.com/item?id=47227791

Points: 1

# Comments: 0

Article URL: https://jonathanlarsen.substack.com/p/us-troops-were-told-iran-war-is-for

Comments URL: https://news.ycombinator.com/item?id=47227762

Points: 32

# Comments: 12

Article URL: https://www.cs.rice.edu/~vardi/comp409/history.pdf

Comments URL: https://news.ycombinator.com/item?id=47227750

Points: 1

# Comments: 0