Loading…
General
Hosted by Unknown Host · General · EN · 0 episodes
Arky is the AI agent deep dive. Every day we take one AI engineering topic and pull it apart in a focused technical interview - sandboxes, agent memory, RAG, evaluation, security, and cost. For each topic we explain the architecture, the trade-offs, and what you'd actually build. Recent deep dives: The Engineering of Sandbox Agents, RAG vs File-Based Memory, You Can't Grade an Agent Like a Model. What you'll hear: - Sandboxing, isolation and agent security - DNS tunneling, microVMs, prompt injection, credential modeling - How agents actually remember - RAG, file-based memory, context management - Agent evaluation and grading - why you can't judge an agent like a model - Lifecycle, state persistence, cost engineering and LLM token economics - Production: local agents vs cloud sandboxes, control planes, snapshots Who it's for: AI engineers, ML practitioners, and technical founders building and operating AI agents. Hosted by Daniel, with technical guest Maya. Full transcripts, show notes, and cited sources for every episode.
Required Pod Score for this show. PitchCentric checks your profile against host openness, topical fit, and audience signals before you generate a pitch.
Contact path
Public web form
Booking probability
36%
Guest openness
Selective
Sign up to generate a grounded pitch for Arky | AI Agent Deep Dive.
Signup to Generate a PitchArky | AI Agent Deep Dive is a general podcast hosted by Unknown Host, with 0 episodes on record and a Required Pod Score of 80. PitchCentric scores this show on Booking Probability, Listen Score, and live audience signals refreshed every 24 hours.
Unknown Host hosts Arky | AI Agent Deep Dive, a general show with 0 episodes published.
Our AI reads these to draft pitches. Use them as grounding for a pitch that cites a real guest and a specific topic.
Episode #5
<p>为什么改动了 AI 助手的配置之后，老任务反而开始失败，模型本身却一点没变？一篇新论文（arXiv 2608.19013）给它起了名字：Harness 层级的遗忘。本集 Daniel 和 Maya 用大白话讲解 Harness 持续学习——冻结模型外层的四个部件（任务接口、经验记忆、能力图谱、自适应路由器），把出主意和上线分开的优化器与评估器（当前改进、历史保持、有效性三道门），以及那个在学得多快和留住多少之间做权衡的旋钮，并给出家庭模拟器、沙盒、推理与视觉任务上的实验结果和诚实的局限。</p>
Episode #4
<p>Why does an AI assistant start failing on old tasks after you change the surrounding setup, even though the model itself never changed? A new paper names it: harness-level forgetting. Daniel and Maya break down Harness Continual Learning (arXiv 2608.19013) - the four harness components around a frozen foundation model, the optimizer and evaluator that gate every change behind current-improvement, historical-retention and validity checks, and the stability-plasticity dial. Evidence from household, sandbox, reasoning and vision tasks, plus the honest caveats of a fresh preprint.</p>
Episode #4
<p>Why does an AI assistant start failing on old tasks after you change the surrounding setup, even though the model itself never changed? A new paper names it: harness-level forgetting. Daniel and Maya break down Harness Continual Learning (arXiv 2608.19013) - the four harness components around a frozen foundation model, the optimizer and evaluator that gate every change behind current-improvement, historical-retention and validity checks, and the stability-plasticity dial that controls the trade-off. Evidence from household, sandbox, reasoning and vision tasks, plus the honest caveats of a fresh preprint.</p>
Episode #3
<p>Daniel and Maya separate the two failure classes of agent sandboxes: the eval-harness headlines (a testing/ops problem, not a boundary defeat) from genuine substrate breakouts like Unit 42's DNS tunneling. They walk through the substrate tradeoff (V8 isolates, microVMs, gVisor), the disposable-agent plus stateful control plane architecture, why attackers chase credentials rather than the model, and why prompt injection is a perception problem no fence can fix.</p><p>Daniel and Maya separate the two failure classes of agent sandboxes: the eval-harness headlines (a testing/ops problem, not a boundary defeat) from genuine substrate breakouts like Unit 42's...
Episode #2
<p>Why you can't grade an agent like a model. The accuracy number we all trusted collapses for agents: a 50x cost gap, reliability that diverges from mean accuracy, a 53-point benchmark swing caused by the scaffold and grader rather than the model, and graders that are an attack surface. Daniel and Maya walk through the fixes.</p>
Sponsor detection runs nightly. Check back soon.
Based on semantic analysis of episode topics and host coverage, this show is a strong guest fit for executives in:
Industry fit is computed by PitchCentric using vector embeddings of the show's episode catalog.
Is this podcast yours and you'd like to remove or correct details? Request removal or email privacy@pitchcentric.com.
FAQs
If you have a concern about deliverability, AI quality, data privacy, or whether this will actually work for your specific situation, it's probably answered below.
Founder Solo gives you 50 AI pitches per month using the credit model (Standard pitches cost 1 credit, Enriched pitches cost 2). Founder Pro raises that to 200 credits per month and adds full Booking Probability access, unlimited Magic Match, Apollo enrichment credits, and data export capabilities. Both plans use the same credit system, so you can stretch your monthly budget further by using Standard-mode drafting.
Agency tiers have no base fee. You pay per managed client and per talent profile. Agency Standard is $199 per client per month; Agency Pro is $399 per client per month. Both add $39 per talent profile per month. Your own team's user seats are always free.
A talent profile represents one person (founder, executive, or spokesperson) you are booking onto podcasts. It includes their bio, topics, headshots, and outreach history. Team plans include 5 profiles; agency plans are pay-as-you-go.
Yes, at any time. Upgrades take effect immediately; downgrades apply at the end of the current billing period. Contact support if you need help migrating between plan families.
Every paid plan includes a 15-day free trial. Your card is saved at signup but you will not be charged until day 16. Cancel any time from your dashboard.
You keep access until the end of your current billing period. No charges after that. Your data is retained for 30 days in case you reactivate.
Yes. Select Annual on the pricing toggle and the discounted price is applied automatically at checkout. The annual price shown is the full year cost.
That is our Enterprise tier. Contact our sales team and we will build a custom plan with volume pricing, a dedicated account manager, and SLA guarantees.
