Case Study · 2024–2026
Six named projects where I’m deploying AI. What each one is, what the AI does inside it, and what it’s producing.
I’m less interested in putting AI on the front of everything than putting it deep in the plumbing — navigating messy data, resources and decisions to make whatever happens on the other side more useful. The front end might be software, an events OS, a marketplace interaction, a community, a real-world experience, a robot entering someone’s home, or an aircraft arriving somewhere nobody expected it. The interesting work is the system underneath and the human experience it makes possible.
What follows is the current state of that thesis as applied across six active projects. For each: what the project is, what we’re actually asking AI to do, and what we’re getting out of it.
Personal operating system · Live at os.mylesweissleder.com
Intent. Keep a complex pursuit pipeline — outreach, applications, warm paths, follow-ups, inbound replies, editorial cadence — legible and actionable, without any of it getting lost in my head or in a dozen tabs. Replace the mental tax of “what’s open, what’s due, who’s owed what” with a surface that answers it.
What the AI does. A three-way AI pipeline runs underneath. GPT discovers and verifies — five parallel scouts across posted jobs, VC funding, inferred gaps, warm network, and regional contract. A second Claude session (“Muse”) reconciles the overnight discoveries against my live state, dedupes against my prior passes, and flags fabricated specifics. The main Claude session operationalizes to my dashboard: writes outreach drafts in my voice, logs sends, patches pipeline bugs in real time. Anthropic’s Haiku runs an overnight pass across the whole corpus to produce a single “north star for today” brief. Happenstance MCP sits in the stack for live LinkedIn-graph queries — semantic warm-path discovery across my 1st and 2nd degree.
Results. 80+ active pursuits tracked with full event history. Six inbox streams (talent-radar, kiln editorial angles, replies, GPT discoveries, GPT signals, Postgres triage) unified into one “new” surface with source attribution and tier labels per item. Gmail API scans every 30 minutes to classify human replies vs auto-mail — 15,000+ calls per month. Warm paths mapped to 24+ named targets at specific companies: DeepMind, Perplexity, Humanoid, Powerdot, Bedrock Robotics. The three-way AI pipeline catches hallucinations in the discovery layer before they reach me (recent example: a scout fabricated a role title and comp band for a company it had warm-signal on, and Muse flagged + killed it before I saw it).
The whole operating-system build is itself the showcase of the thesis. The AI isn’t in the interface. It’s in the plumbing — discovery, reconciliation, ranking, voice calibration, pipeline health. What I see is a quiet dashboard with the right next action already surfaced.
Global events program · 2026 · Four continents
Intent. Run a forty-plus event portfolio for an electric aircraft company introducing itself to the world — flight demonstrations, international air shows, consumer activations at cultural moments, trade shows, partnership deployments with Uber, Blade, Virgin Atlantic, Toyota. Four continents in six months. A distributed team of planners, outside vendors, and me producing specific moments on the ground.
What the AI does. Not in the attendee interface — where it would feel like a gimmick. In the production layer: schedule reasoning across time zones and vendors, document normalization across different venues’ permits and insurance frameworks, lift-and-shift of show-flow logic from one event to another, intake summarization on partner briefs. Mostly workflow translation and context-handoff work, so my planners can get a running start on each new event rather than rebuilding the brief every time.
Results. Four marquee events produced and managed end-to-end on-site, back-to-back, across the peak 2026 summer window — US Open at Shinnecock, FIFA Fan Zone at USTA Queens, National Mall for America 250, Potters Field in London with Virgin Atlantic livery and Tower Bridge in the backdrop. The London arrival generated 77 pieces of tracked press coverage across UK national dailies, business titles, and international pickup. As Joby’s CEO has publicly framed it, AI is transforming “everything from aerodynamics research to back-office tasks” across the company. The events function wasn’t scrambling to shop for tools when leadership pointed at AI — the scaffolding was already there.
(Deeper story: Events OS case study →)
Director of Communications · AAMI · 2026
Intent. Thought-leadership cadence across the Advanced Air Mobility space — drones, electric aircraft, autonomous flight — held to an ethical, operator-centric editorial doctrine. Named companies as the eyecatcher, business analysis as the substance, trust-earning tone, avoid becoming a cheerleader for any one OEM.
What the AI does. “Kiln” is a scheduled editorial job that generates three angles per day against the current industry signal set, held to the doctrine. Draws from recent funding events, regulatory news, deployment milestones, and industry quietude. Each angle lands in an inbox I review and edit; the ones I ship go out under my byline, voice-calibrated against a curated exemplar corpus of my prior writing. The AI handles research scaffolding and voice drift checks; the editorial judgment and sentence-level writing stay mine. Published pieces include an attribution line acknowledging the AI-assisted workflow once — via a static editorial-process page — not per piece, so the byline standard holds.
Results. 3 angles per day for human review, averaging roughly one piece per week that makes it through to publication under AAMI framing. Doctrine survives the AI layer: no Joby-centric pieces (per the voice policy, since Joby is a former employer), named-company analysis where warranted, no greenwashing. The AI mostly earns its keep by killing the writer’s block of “what should I even write about this week” and freeing the editorial hours for the hard part — the actual argument.
Shipping consumer product · yesavailable.link · 2025–2026
Intent. A peer-to-peer sales commitment layer — not a marketplace replacement, a trust layer for the moment when a casual seller and a serious buyer need to close. Target seller is a casual, episodic poster (2 to 6 purges a year, 3 to 8 items each time) who hates chasing flakes on Facebook Marketplace, Nextdoor, and Craigslist. Platform optimizes the “is this still available?” moment.
What the AI does. Casual sellers upload photos and a few bullets. AI drafts the item description in a plain-spoken voice, suggests pricing from comparables, writes the share copy tailored to each channel (Craigslist is URL-prohibited; Nextdoor has an authorized smart-link API; Facebook Marketplace is grey territory). Incoming buyer messages across SMS, email, and web replies get normalized and ranked for intent; serious holds surface at the top, flake-signal replies drop. None of this is marketed as “AI.” The copy on the seller surfaces says “we draft it,” not “AI drafts it,” because the user doesn’t care about the mechanism.
Results. Product is live at yesavailable.link, deliberately in a muted go-to-market posture during an active job search. First real seller stress-tests have validated the thesis that casual sellers want their descriptions written for them more than they want “AI” branding on the surface. Underneath, the platform is a laboratory for the broader thesis: AI deep in the plumbing of decisions and messy data, not on the front of the user experience.
Head of Community-Led Growth · Pivotal · 2023–2025
Intent. Build trust around personal electric flight at the first operator-ready eVTOL company shipping its Helix aircraft. Not aviation-industry trust — actual human-walks-up-to-the-aircraft trust, with people whose relationship to flight is a middle seat on a Boeing. Community-first GTM in a category where the audience didn’t yet exist.
What the AI does. Lead qualification across inbound interest — airshow visitors who scanned a QR code, Facebook group members, demo-day attendees — routed through a model that classified interest intensity, geographic fit, and likelihood to become either a buyer or an amplifier. Community-signal mining across the enthusiast forums where personal-flight conversations were already happening. Trip-planning for the deployment calendar — which cities, which cultural moments, which demo surfaces. AI as a researcher behind the operator decisions, never writing to the community directly.
Results. Community-led channel generated approximately 25% of new leads during peak deployment windows — a reduction of about 34% in cost-per-MQL relative to paid-digital benchmarks. (Deeper story: Pivotal case study →)
Fractional practice · thehighertide.com · 2025–Present
Intent. A fractional-to-embedded commercialization practice for frontier-tech founders who are post-engineering, pre-“we know how this reaches humans.” Humanoid robotics, consumer EV, personal flight, autonomous systems, physical AI. Short engagements that leave a working operating layer behind, not a slide deck.
What the AI does. AI runs the warm-path mapping for pursuit: every target company gets a Happenstance query across my LinkedIn graph before any cold-outreach draft gets written. AI drafts every outreach piece calibrated to my voice corpus, with the specific research signal that triggered the outreach surfaced in the opening. AI prepares pre-call dossiers so I walk into first conversations with real context on the founder, the stage, the competitive landscape, and the recent shifts in their world. The AI never sends — every send is mine, at human cadence, with voice intact.
Results. Active pipeline across founder-level conversations with Bedrock Robotics (post $600M Series C), Humanoid ($152M Series A, 34,000 pre-orders), Perplexity (Director of Events role), Pete Huang at XDOF (Series A $70M), and warm reconnects across the Pivotal and SF New Tech alumni networks. Over 70% of outreach lands in-category and in-register because the AI handles the pattern-matching layer that would otherwise eat an entire week of research before a single email went out.
Across six projects, the AI is doing the same kind of work in each one: handling the pattern-matching, the normalization, the schedule reasoning, the research scaffolding, and the voice calibration — so that the human judgment and the sentence-level decisions and the on-the-ground work still happen at human cadence, with my hands on them. The AI is in the plumbing, and the plumbing matters, and I’m the kind of operator who builds the plumbing before someone else asks me to.
If any of this is relevant to what you’re building — frontier technology that needs to meet actual humans, with an AI-native operating layer underneath — find me.