The Workbench — Launch Plan
Prepared June 5, 2026 · Target launch: week of July 6, 2026 · Cadence: one episode every two weeks
1. Assessment of the concept
What is strong. The project has a rare combination: a working tool, real teaching material, and a credible first-person AI story. The “make the machinery legible” framing already on the About page is excellent — it unifies the finance stream and the LLM stream under one idea, and it is honest about your position (educator-experimenter, not AGI guru). That honesty is a differentiator in a space crowded with hype. The three LLM essays already published are substantial and give the blog immediate credibility.
The main risk: scope. Four streams × three channels × an episode bundle of model + video + long post + newsletter is a publishing house’s workload, not one professor’s. At one episode per two weeks, you produce ~6 episodes by end of September. The plan below assumes that budget and cuts accordingly.
Three structural recommendations:
Make the blog the canonical home. Every episode lives fully on the blog. The newsletter is a digest + link; the video is a companion, embedded in the post. This means one piece of “real” writing per episode, not three. Substack readers lose nothing (you promised equal depth for free anyway) and you avoid maintaining parallel long-form content.
Lead with Stream 4 (AI-assisted financial analysis), not Stream 2. You chose “mixed audience, lead with AI angle” — Stream 4 is that angle: finance content with the AI hook visible in every episode. Stream 2 (finance class basics) works better as a supporting reference series that Stream 4 episodes link back to. Stream 3 (ABC of language models) is already alive on the blog and continues as the second track.
Defer the pipeline. Portfolio optimization, crypto simulation, fuzzy systems: park them. Mention them once in the manifesto as “future workshop projects” and stop there. Announcing too much creates debt.
Suggested stream priority for the first 12 weeks:
| Priority | Stream | Role |
|---|---|---|
| 1 | AI-assisted financial analysis (Stream 4) | Flagship — the differentiator |
| 2 | ABC of language models (Stream 3) | Already running; alternates with Stream 4 |
| 3 | My finance class (Stream 2) | Starts in September with the academic year |
| — | Everything else | Future projects, unannounced |
Alternating Streams 4 and 3 every two weeks gives each a monthly rhythm — sustainable, and each serves a different half of your mixed audience.
2. Inauguration day (week of July 6)
Launch with welcome content plus exactly one flagship episode. A manifesto alone is a promise; a manifesto plus a working demonstration is proof. More than one episode dilutes attention and burns backlog.
On the day:
- Blog: the Welcome/manifesto post (draft provided separately), pinned. The three LLM essays remain as existing depth.
- YouTube: channel goes live with two videos — a 90–120 second channel trailer (the manifesto, spoken, over screen captures of the app) and Episode 1 of Stream 4: an AI-assisted financial-analysis session on the financial statement analysis model — assumptions in, formulas, calculation, error correction, report out, driven through an AI coding assistant.
- Substack: Welcome issue (draft provided separately) announcing the blog, the channel, and Episode 1, with the episode summary and links.
- Cross-links everywhere: blog header links to YouTube and Substack; YouTube channel description and every video description link to the blog post; Substack About page links to both.
Why Episode 1 from Stream 4: it shows the tool, the AI workflow, and the finance substance in one artifact; it is the episode only you can make. The LLM essays already cover the launch for the AI-curious reader.
Pre-launch production checklist (4 weeks):
- Week 1: finalize manifesto post and Substack welcome; create YouTube channel (see §3); script Episode 1.
- Week 2: record and edit the channel trailer; record Episode 1 demo takes.
- Week 3: edit Episode 1 video; write Episode 1 blog post; prepare the episode’s model file/screenshots.
- Week 4: buffer — captions, thumbnails, cross-links, dry-run of the whole chain; schedule everything.
If Week 4 is consumed by overruns, slip the date rather than the quality. Also begin Episode 2 (Stream 3, a tiny-Transformer video walking through one of the trained dialogue models) so you have one episode of runway at launch.
3. YouTube channel setup
(Complements the technical advice you saved in the earlier chat — this covers format and policy.)
- Identity: name the channel The Workbench to match the blog; use the same wordmark/artwork. Link it to your Google account as a Brand Account (not the bare personal account) so the channel name is the brand and you can add managers later. Custom handle:
@theworkbenchor nearest available (e.g.@lerzegov-workbench). - Format: screen capture of the app + your voiceover; brief on-camera or still intro is enough — the screen is the star. Record at 1080p minimum; keep UI text legible (zoom the app, don’t shrink the capture).
- Duration: 8–15 minutes for episode videos. Long enough for substance, short enough to finish. Anything longer belongs in the blog post. The trailer: under 2 minutes.
- Structure per video: 30-second cold open showing the end result → what we’ll build → the demonstration → recap + “full write-up on the blog” pointer.
- Per-video hygiene: descriptive title (problem first, tool second: “Forecasting an income statement with an AI assistant” beats “Impromptu demo #1”); chapters/timestamps; link to the blog post first in the description; auto-captions reviewed and corrected (accent + technical vocabulary).
- Comments: on, with moderation queue for links. Answering early comments personally is the cheapest community building available.
- Shorts: skip for now. One more format is one too many at launch. Revisit after ~6 episodes by cutting 60-second highlights from existing videos.
4. Making the app available: decision
Recommendation: show, don’t ship — for now. At launch, the app appears in videos and posts but is not downloadable. This is the standard early-access pattern and nobody will find it strange; a “request early access” path makes it a feature (“hands-on sessions with the author”) rather than a withholding.
Reasons it is the right call, not just the cautious one:
- Support load. Public downloads create an implicit support contract. A two-tier app with a server backend, DB and data-provider integrations is not a double-click install; early public users would mostly generate installation questions, not modeling feedback.
- License and IP are unresolved. You are considering MIT later; shipping binaries now under an improvised license creates ambiguity you’d have to walk back. Also worth checking quietly before any public code release: the data-provider integration (LSEG terms) and anything carrying obligations from the academic alliance around your prior tooling.
- Your stated constraint. Open-sourcing without a technical leader risks an unmaintained public repo, which signals worse than a private one.
Staged path:
| Stage | When | What |
|---|---|---|
| 1. Show only | Launch | App in videos/posts; “early access — get in touch” note on the About page |
| 2. Invited early users | From ~Sept | 5–10 hand-picked users (colleagues, students, a couple of practitioners from your audience), private repo access or distributed builds, a simple email/group channel for feedback, explicit “as-is, at your own risk” note |
| 3. Public pre-release | When stable + conditions met | MIT license, public repo, install docs — only if the community/leadership condition from your notes is satisfied; otherwise stay at Stage 2 indefinitely, which is a perfectly respectable steady state |
Documentation now: only what early users need — a getting-started page and a model-format overview, in the private repo. Public tech docs belong to Stage 3.
Interaction with the audience meanwhile: blog comments are not needed; use Substack comments + YouTube comments + email. That is enough surface for a solo operator.
5. Twelve-week roadmap
| Week of | Action |
|---|---|
| Jun 8 – Jul 3 | Production (checklist in §2); create channel; finish cover content |
| Jul 6 | Launch: manifesto + trailer + Episode 1 (AI-assisted analysis) + Substack welcome |
| Jul 20 | Episode 2 — ABC of LMs: inside a tiny dialogue Transformer (video companion to the existing essays) |
| Aug 3 | Episode 3 — AI-assisted analysis: report generation and error correction |
| Aug 17 | Episode 4 — ABC of LMs: training the tiny model, watching it learn |
| Aug 31 | Light week (field note only) — build September backlog |
| Sep 14 | Episode 5 — My finance class opener: forecast income statement (academic year tie-in); start inviting early users (Stage 2) |
| Sep 28 | Episode 6 — My finance class: capital budgeting · then review metrics and adjust |
The one metric to watch per channel: blog — returning readers; YouTube — average view duration (are people finishing?); Substack — subscriber growth. Ignore everything else for six months.
6. Priorities, condensed
Do now: manifesto + Substack welcome (drafted), channel creation, Episode 1 production. Do soon: Episode 2 so launch has runway. Decide later (September): early-user invitations, finance-class stream start. Don’t do yet: public downloads, license choice, Shorts, comment systems, the pipeline topics.