Shoreline Ep. 1

The gist

Episode 1 is really about the moment AI stops being a clever writing tool and starts becoming a business operating layer. The hosts keep circling the same practical question: what context should an agent see, what should it be allowed to change, and where should the human review happen?

The answer is not one tool. It is a system: local models for experiments, cloud models for hard reasoning, agents with narrow lanes, shared memory in Notion or a CRM, and a review loop that catches the messy parts before they touch customers or revenue data.

Key takeaways

Five things to steal from this episode.

01

Local AI is a fallback lane.

A Mac mini with Ollama is useful for privacy, offline experiments, and cheap iteration. It is not automatically better than frontier cloud models.

02

The agent stack is context plumbing.

Hermes, OpenClaw, terminals, Telegram, MCPs, and APIs only work when each agent has a clear lane and a clear source of truth.

03

Memory needs backups.

A company brain becomes fragile when agents can bulk-edit records. Backups and approval gates matter before the first destructive workflow.

04

Review is the bottleneck.

More agents create more output. The system needs statuses, owners, summaries, changed-record logs, and one place to inspect the work.

05

Taste does not get automated away.

The best workflows use AI to compress blank-page work while keeping human judgment around tone, usefulness, risk, and product quality.

Tactical guidesWhat to try next

Guide 01

Set up a local model lab.

Use local AI for offline fallback, private drafts, repeatable tests, and learning how model quality feels at different sizes.

Pick one always-on machine, such as a Mac mini or spare laptop.
Install Ollama and leave enough disk space for model files.
Start with one small model family and test the same prompt against cloud models.
Use local models for low-risk work: summaries, drafts, checklists, and automation tests.
Escalate to a frontier model when the task needs deep reasoning or current web context.

Guide 02

Build an agent bench before an agent army.

Start with named roles instead of spinning up more surfaces than you can review.

Name each agent and write its job in one sentence.
Assign one primary surface: terminal, Telegram, Slack, browser, or desktop.
Give each agent one lane: CRM cleanup, analytics, research, or inbox triage.
Require a daily review note with changes, outputs, and unresolved questions.
Retire workflows that feel impressive but do not create reviewed value.

Guide 03

Create a context and permissions map.

Most friction comes from reconnecting analytics, ads, files, skills, MCPs, APIs, and local folders across machines.

List every system an agent needs to read, write, or never touch.
Split access into read-only, draft-only, and approved-write permissions.
Document which MCP servers, APIs, keys, and folders power each workflow.
Keep credentials outside prompts and store setup steps in project docs.
Run a dry test before live write access to customer or revenue data.

Guide 04

Turn Notion and CRM data into a company brain.

The strongest business workflow is simple: capture calls, clean records, preserve tags, and make the next action obvious.

Choose a source of truth for contacts, companies, notes, tasks, and transcripts.
Define required fields before an agent can update a record.
Back up the database before letting agents perform bulk edits.
Use agents to draft field updates, next steps, and daily work queues.
Review destructive changes, especially tags, owners, and statuses.

Guide 05

Use AI to learn without outsourcing taste.

The show connects book notes, founder clips, and personal libraries as ways to absorb ideas more deeply.

Collect books, articles, transcripts, and clips that already matter to you.
Ask an AI assistant to create summaries, themes, and retrieval tags.
Query the library with real situations: a sales problem, design decision, or training plan.
Return to the original source when the answer matters.
Publish or teach the idea back to force active understanding.

Guide 06

Build a review loop for token-heavy work.

Running many agents creates more output than one person can casually inspect. Design the review surface first.

Give every agent output a destination, owner, and due date.
Require summaries with changed files, changed records, and open questions.
Mark outputs as experiment, draft, reviewed, or shipped.
Compare agents on the same task only when the learning is worth the review cost.
Use failures as data: what permission, context, or instruction was missing?

ChaptersTimestamp map