Nathan Weill here.

I'm the CEO @ Flow Digital, where we help companies unleash their full potential by strategically automating every inch of their workflows. Each week, I share hot AI and automation tips to help you move your business into the future successfully.

Tools to try (today)

A CRM That Keeps Every Team on the Same Page

Clarify fills in its own records, gives every team a login, and is SOC 2 Type 2 compliant.

Sales reps spend just 28% of their week selling. When updates wait on someone typing them in, sales, customer success, and finance end up working from different pictures of the same customer. Clarify fills itself in, so every team starts from the same record.

  • Records that update themselves: connect your email and calendar, and calls, meeting notes, and attached contracts and decks land on the right person, company, and deal.

  • Handoffs with nothing missing: require a field like contract value before a deal can move to Won, so the next team starts with what was promised.

  • A login for every team, with limits you set: every plan includes unlimited seats, including the free one, and admins can restrict what each person sees down to individual records.

Customer success and finance can see the same deal history without asking sales for it. Clarify is also SOC 2 Type 2 compliant, an independent audit of how it protects customer data, and you can request the full report.

It suits small teams best. Clarify describes its built-in integrations as limited, so connect your finance and customer success tools with Zapier.

Get more out of

Softr: Build Interactive Content Without a Developer

Gif of an interactive checklist built with Softr

Use Softr to build interactive content any team can ship in an afternoon

Static PDFs and landing pages get skimmed. Interactive tools get used, but most teams need a developer to build one.

Softr is best known for client portals, internal tools, and CRMs built on top of your existing data. Its Vibe Coding block adds a faster path: describe the page you want, and it builds and hosts it.

A non-technical teammate built the Smart Buyer's Guide to Choosing an AI Expert, a live interactive page, with no developer involved.

What sets it apart:

  • Hosting included: your page goes live on Softr's infrastructure, with no servers to set up.

  • Your data, connected: pull from Airtable, Notion, or Softr's own database so content stays current.

  • Undo built in: block history rolls back any AI change you don't like.

  • Brand and access rules inherited: pages match your app's theme and respect user permissions.

Best practices:

  • Give the page one job: help a buyer decide, qualify a lead, or score a self-assessment.

  • Describe the outcome you want, then refine one change at a time.

  • Keep content in a table, so updates need no rebuild.

  • Check it on a phone before you share it.

Teams that can describe the page they need can now ship it themselves.

A better, faster, smarter way to

Cut Your AI Bill: Match the Model to the Job

The cheapest tier costs up to 100 times less than the flagship, and plenty of daily work doesn't need the flagship.

Every task you send to a flagship AI model costs more than it needs to. Anthropic's cheapest Claude model costs $0.10 per million input tokens. Its top tier costs $10. That is a 100x gap, and a lot of daily work (sorting emails, pulling fields from invoices, summarizing a thread) does not need the top tier.

The fix is a habit. Sort your tasks into three tiers, and start each one on the cheapest tier that could do the job. OpenAI, Anthropic, and Google all sell the same three.

  • Small tier: sorting, extracting, routing, and short summaries

  • Mid tier: drafting, analysis, and most daily work

  • Flagship: multi-step reasoning, complex code, and high-stakes decisions

Move up a tier only when the output falls short. For work that can wait overnight, Anthropic and Google both cut the price 50% with batch processing.

ChatGPT (OpenAI)

  • Small tier: GPT-6 Luna, $0.10 in / $0.50 out per million tokens

  • Mid tier: GPT-6.1 Sol, $2 / $10

  • Flagship: GPT-6 Astra, $10 / $50

OpenAI lists all three on its API models page. In the ChatGPT app, the picker labels change often, so use the default for everyday work and switch to a higher reasoning setting only when an answer comes back thin.

Claude (Anthropic)

  • Small tier: Haiku 5.5, $0.10 / $0.50

  • Mid tier: Sonnet 5.5, $2 / $10

  • Step up: Opus 5.5, $4 / $20

  • Flagship: Fable 5.1, $10 / $50

Anthropic's own model selection guide suggests Opus 5.5 as the starting point for demanding work, and Fable 5.1 when your tests still fall short. For the everyday tasks above, start with Haiku or Sonnet and step up only when the output needs it. Cached prompts cost a fraction of the base input price, which helps when you send the same long instructions over and over.

Gemini (Google)

  • Small tier: Gemini 3.5 Flash-Lite, $0.30 / $2.50

  • Mid tier: Gemini 3.8 Flash, $0.75 / $3.75 through December 31, then $1.50 / $7.50

  • Flagship: Gemini 4 Argon, $2 / $10 through November 30, then $4 / $20

Google lists every current model on its Gemini models page. Argon's price doubles on December 1, so lock in your testing and budgeting before then.

Where to look to stay current

Model names and prices turned over several times this year. Check these before you renew a plan or start a big project:

  • The three vendor pages above: the source of truth for names, prices, and retirement dates

  • Anthropic's deprecations page: shows how long a model is guaranteed to stay available

  • Google's pricing page: flags introductory rates before they expire

  • Artificial Analysis: compares models on intelligence, speed, and cost per task, and its model recommender narrows the list to your priorities

  • LMArena: shows how people rate models side by side on real prompts

Build a ten-task test

Pick ten real tasks from your own work, such as a customer reply, an invoice to extract, and a report to summarize. When a new model launches, run all ten on the cheaper tier first and compare the results. If the small tier passes, you just cut that workflow's cost, and a quarterly calendar reminder keeps the habit going.

Prices checked October 2026.

Insider insights

Why Big Companies Are Paying to Embed AI Experts in Their Teams

How the "FDE" role works, and the lessons all companies can apply

Most AI projects stall after the demo. The tool gets bought, the pilot impresses everyone, and the daily work stays exactly as it was. Large companies now pay a premium to fix that with one hire: the forward deployed engineer, or FDE.

An FDE works inside your team instead of at a vendor's office. They learn how the work really runs, then build and adjust the AI workflow until people use it every day. The role started at Palantir, and U.S. postings have grown from fewer than 200 in 2023 to more than 7,300 in the first nine months of 2026, according to labor-market firm Lightcast.

The price tag explains the rush. Advertised pay has a median near $195,000 a year, before benefits and recruiting costs, and big companies are paying it. Smaller businesses do not have to match that spend. Our team includes forward deployed engineers who work inside a client's process in smaller, scoped chunks, from a single workflow to a short embedded sprint.

Whatever your size, the habits behind the role are worth copying.

Four Habits That Make AI Stick Without the Headcount

You may never hire an FDE, and you can still get the result. The role works because of how it operates, and each of those habits transfers to a team of any size.

  • Start with one workflow you can watch: pick a single process, like quote follow-up or invoice reminders, and sit beside the person who runs it for a week. You will find the real steps, the workarounds, and the delays that no demo covers.

  • Give one person ownership after launch: name someone on staff, even for a few hours a week, to keep the automation working and fix it when it breaks. An automation with no owner gets abandoned the first time something changes.

  • Decide where a person stays in the loop: write down the calls that stay human, such as refunds, pricing exceptions, and anything a customer might escalate. Everything else can run on its own.

  • Bring in an embedded engineer for a defined stretch: have one work inside a single process for a few weeks, then hand it back with documentation and an owner who knows how it runs. You get the FDE approach at project cost instead of a salary.

Our team does this work at the size that fits: one workflow, a few weeks embedded, or a longer engagement. If you want to start on your own, the first two habits cost attention and nothing else.

Want a second set of eyes on where an embedded AI expert would help most? Talk with our team about one workflow, and we will point to the first place it pays off.

SMART WORDS OF THE WEEK:

"What I cannot create, I do not understand.”

— Richard Feynman

AI now builds the page or the workflow from your description, so few of us read every line it writes. What you still have to understand is the job itself: what it should do, who owns it, and how you will know it works.

Nathan Weill

CEO

Stay in touch with our team from anywhere.