Writing
Practical breakdowns from building AI systems in production, what actually works, what quietly breaks, and how to take a clever demo all the way to dependable production.
Outbound is now a systems problem, not a headcount one. The full stack: target and qualify, decide build vs buy, personalize deeply, reach across channels, and aim it all at the accounts that already show intent.
Templated variables fool no one. Real personalization, the kind that gets 5-10% reply rates, researches each lead deeply and writes from it. How to do that at scale, and the prompt craft that makes it read human.
A LinkedIn note, then an email that references it, then a call moments after a positive reply, orchestrated as one conditional flow. How multi-channel outbound agents work, and how to use them responsibly.
Demoing an AI product is easy; shipping one that survives production is a craft. The disciplines that matter: when to use an agent, engineering context, applying product judgment, and designing for reliability.
The most interesting edge of AI is where it meets money: agents that pay on their own, the economy forming around them, and the quant discipline that turns markets into math. A map, from someone who spent years in both.
No public benchmark can tell you which model fits your task, a tiny private eval can. How to choose: optimize cost per successful outcome, use the thinking and effort dials, and shift the curve with caching and context hygiene.
Claude Code launched a year ago to a muted reaction. Now people direct fleets of agents, sometimes from their phones. The two big leaps, the habits that stuck, and what it means for how we work.
HTTP 402 sat unused since 1997. Now Coinbase's x402 protocol uses it to let any API or website charge per request in stablecoins, the payment rail for the agent economy. What it is, how it works, and where it stands.
Once an AI agent can pay on its own, it becomes an economic actor, not just an assistant. Why the agent economy runs on stablecoins, the three layers it needs (identity, payments, reputation), and what to build.
Quants don't draw trend lines, they quantify the market and size bets by probability. How hedge-fund regime models actually work, the rigor that separates edge from mirage, and what AI changes.
The demo always works; production is where agents quietly fail. The five failure modes I see most, context rot, subagent sprawl, no verification, no evals, unbounded autonomy, and how to design around each.
The complete guide to AI search visibility (GEO/AEO): what it is, how ChatGPT, Perplexity, and AI Overviews differ, how to get cited, and how to measure your share of voice. Your map to becoming the answer.
A practitioner's guide to building real software with Claude Code: the working habits, the agent architecture (tool vs skill vs subagent), and the prompting discipline that make it reliable.
Anthropic's guidance on effective agents in three ideas: don't build agents for everything, keep them simple, and think like your agent. What each one means in practice.
As models got better at following instructions, the hard part moved from wording the prompt to assembling the right context. What context engineering is, its building blocks, and why minimal beats maximal.
Where AI gives a product manager real leverage today (research, insights, data, prototyping) and where to keep human judgment (vision and the strategy bet), plus the one prompting tip that matters most.
Clay is excellent, and credit-metered. Here's the build-vs-buy framework for your outbound data and enrichment stack: when Clay's speed wins, when building your own with agents wins, and the lever that matters more than either.
When an agent's prompt hits 400 lines and performance degrades, the fix isn't a smarter model, it's the right primitive. A decision framework for tools, skills, and subagents, drawn from Anthropic's own engineers.
Prompt like an engineer: start from evals, clean up before you optimize, give tools instead of pleas, state both sides of every trade-off, and decompose hard tasks into a generate-evaluate-repair loop.
AEO and GEO are the same thing, SEO isn't dead, and the real change is narrower than the hype. What genuinely changed in AI search, what carries straight over from SEO, and how to invest.
The people who built Claude Code don't out-prompt everyone. They fix the system instead of the instance, make verification agent-native, and run fleets of agents on auto mode. Here's what actually transfers to how you build.
AI search isn't one channel. Only ~14% of top-cited domains overlap across the three engines. Here's what ChatGPT, Perplexity, AI Overviews, and Google's AI Mode each trust, and how to win citations on each.
Answer engine optimization, explained by someone who measures it daily: how AI answers actually pick sources, why being mentioned beats ranking #1, and the playbook to get cited in ChatGPT, Perplexity, and AI Overviews.
Your analytics can't see most AI traffic. Here's how to actually measure AI search visibility: share of voice, per-engine tracking, and why you have to measure it probabilistically.
The playbook for building an AI SDR team that runs cold outbound on autopilot, from ICP to inbox, and the product foundations most teams skip.