TribalScale

Healthcare AI has no soft launch. A tool that records patient conversations either meets HIPAA on the day it ships or it does not ship.

Skriber partnered with TribalScale to build clinical documentation where compliance was a build constraint from sprint one. Vanta was integrated at the start, so HIPAA readiness moved in parallel with the product instead of becoming a gate at the end. Real-time capture runs on Deepgram. Summaries run on OpenAI GPT and Groq.

That sequencing shaped the product. Session history ships with edit and delete because clinicians need to correct the record. Manual notes sit alongside AI summaries because the model does not get the last word on a patient chart. Phase two added the native app and Stripe payments on the same compliance footing.

Ambient scribes are now a crowded category. What still separates a pilot from a product is whether the compliance posture was designed in or bolted on. Skriber shipped on the first footing, and the platform has grown into ambient documentation for clinicians since.

🔗 The full build, stack, and phasing: https://hubs.la/Q04x93HV0

For twenty years, search sent media companies their audience. Traffic sold the ads. The page was the product.

That arrangement has come apart, and Chartbeat has measured how far. Over two years:

→ Small publishers lost 60% of their search referral traffic
→ Mid-sized publishers lost 47%
→ The largest lost 22%

The people running these companies do not expect a floor. The 280 news executives the Reuters Institute surveyed across 51 countries expect their own search traffic to fall another 43% over the next three years.

No channel has taken its place. Chatbots account for under 1% of publisher page view referrals, and ChatGPT referrals grew more than 200% year over year while still sitting at 0.02% of total referral traffic. Licensing has not filled the hole either. In that same survey, no publisher named AI platform deals as its expected main source of income.

With no channel to re-optimize for, the companies with a path forward are building revenue they control:

→ Subscription and membership, the top revenue focus for 76% of commercial publishers surveyed
→ Events, at 54% and rising for three years
→ USA Today Co.'s other digital revenue (syndication, affiliate, AI partnerships), up 126% year over year to $33.75 million in Q1 2026

Every item on that list is a product with a build cost behind it: payment infrastructure, subscriber data, ticketing and CRM, content pipelines that feed partners. That work is venture-building inside an operating business, on a deadline, while the old model still pays the bills.

TribalScale builds and ships digital products with media companies: the platforms, data foundations, and subscriber experiences that turn an audience into revenue you own.

Swipe through for the numbers.

Anthropic announced Enterprise Frontier Safeguards on September 1. Detecting serious misuse means holding data long enough to correlate behaviour across sessions and accounts, and regulated enterprises cannot leave that data with a vendor. EFS puts monitoring data in the customer's own cloud account, under their own encryption keys, access policies, and audit logging, with flags routed to their security team.

Four questions for any enterprise AI deployment:

→ Where does the data live, and who holds the keys?
→ Who owns escalation when a flag fires?
→ Can monitoring see across sessions, or one prompt at a time?
→ Which controls are optional, and who decides?

Data residency, key custody, and review authority get decided early and are expensive to reverse. Worth answering before the pilot, not after.

AI assistants are easy to add. An enterprise agent platform is harder to build.

TribalScale’s case study shows what changes when disconnected automations are brought into one governed architecture:
→ Shared agent runtime and enterprise memory
→ Task-specific tool access
→ Auditability, privacy controls, and human review

The result is a production-scale multi-agent ecosystem supporting delivery operations, People & Culture, proposal development, case-study generation, and employee productivity.

Reported outcomes include case-study creation falling from days to approximately 15 minutes, 20 employees adopting Personal Agents, and enterprise-scale operations at approximately $20–40 USD per day.

The lesson: moving from pilots to production is an architecture and governance problem, not just a model-selection problem.

Banking made more net income than any other industry last year. It also still can't get people through an account application.

For every checking account a bank opens online, 3.36 applications get abandoned. That ratio has barely moved year over year, and digital is the growth channel, accounting for 27% of new checking accounts.

Meanwhile 34% of retail banking customers say they switched providers to get a better website or mobile app. Pricing and rates ranked lower.

Forrester's 2026 rankings say the same thing from the other direction: no US bank earned an "excellent" score on customer experience or brand experience. Scores rose for the first time in years. Most experiences are still mediocre.

Once we're inside these systems, intent is rarely the constraint, and budget usually isn't either. The application flow sits on top of data that arrives in batches, identity and fraud checks tuned for a branch, and a core that was never designed to answer a question in 200 milliseconds. A quarter of applications get rejected by those fraud controls before a human ever looks at them.

You can't design your way out of that in Figma.

Fixing the product layer means fixing what the product layer stands on.

Swipe through for the numbers.

Salesforce and Anthropic's Claudeforce announcement is less a model story than an architecture story.

The companies say Salesforce in Claude brings live CRM context, business rules, permissions, and governed actions into Claude, starting with 37 prebuilt sales skills. Select pilots are available now; an open beta is expected in September.

What has to be true before an agent workflow reaches production:
→ Context and identity
→ Least-privilege access
→ Rules and review
→ Audit and a measurable KPI

The interface may change. The delivery discipline does not. Start with the workflow and outcome, then design the data, systems, controls, owners, and adoption path around it.

Your CRM ships an MCP server. So does your knowledge base, your cloud provider, and most of the SaaS in your stack.

Which means the integration question has moved. Not whether to standardize, but which of your systems an AI agent should be able to reach, and who signs off on that.

TribalScale's recommendation is to start with an audit rather than a protocol decision:
→ Catalog the data sources your AI initiatives actually need
→ Calculate what integration and maintenance cost you today
→ Map those constraints against your AI roadmap
→ Decide where standardization helps and where it widens a permission surface

An MCP server is a thin adapter over an API you already own. It changes who can reach that system and how easily, which makes it a governance decision before it is an architecture one.

Read the full insight: https://hubs.la/Q04vy6Z_0

Most employees will never become great prompters, and our CEO Sheetal Jaitly argues they should not have to.

A licence for every employee creates individual productivity gains. It does not change how the business operates. The system should capture information once, route it where it needs to go, automate the repeatable actions, and surface the exceptions that need human judgment.

AI has also changed the build versus buy calculation. The workflows that create differentiation deserve a different answer today than they got two years ago. Which ones those are is the harder question: https://hubs.la/Q04vnLC60

Manufacturers aren't debating digital transformation anymore. 90% now call it essential to staying competitive, and COOs plan to keep raising AI budgets over the next five years.

But scaling it is a different problem. Only 2% of companies report AI fully embedded across all operational functions. Nearly half of surveyed COOs point to the same root cause: data and IT/OT systems that weren't built for this.

The pattern holds across the sector — most manufacturers focusing AI on operations are still piloting it, not running it in production.

The budgets are there. The systems aren't yet. Swipe through this week's numbers.

Enterprise buyers are changing what they'll sign for.

Reuters reports about 80% of TCS's business services contracts, meaning finance, HR, and back-office work, are now tied to performance measures, per CEO K Krithivasan. That's roughly double where it sat in late 2023, according to the report.

The two smaller details in the piece stack on top of it. Persistent Systems CEO Sandeep Kalra says clients want the same work at 25 to 30 percent less, delivered faster. And contracts are getting shorter, with some work moving in-house rather than out to a vendor at all.

Less money, less time, more conditions. Three signals that buyers are hedging until the cost of AI delivery settles.

Which of the three is showing up in your contracts?