What moved in tech this week — and what it means for EdgePointe
Week of September 28, 2026 · a 5–7 minute catch-up
Nine cards, ~30 seconds each. You get the so-what; Paul carries the detail. Hit Play narrated to listen.
EDGEPOINTE · VANTAGE
The big picture · our bet
Three-for-three in five days: the builders took the generic half of our pitch Bet narrower
Our bet has never been on which AI company wins. It is that somebody operates this for a small business, using that business's own material. This week tested that hard.
Wix
A built-in AI phone agent on any paid plan. The generic receptionist is now a checkbox.
Webflow
An agent can now build, animate and deploy a whole site inside their editor.
Squarespace
AI multi-page sites plus Image Studio — reshoots photos and builds matched image sets.
None of the three starts from a specific venue's leads, calls, tours, photos and reviews, and none of them runs the site afterward. That is what is left, and it is ours.
Bottom line: our edge went from "we can build that" to "we start from your material and we run it." Every pitch needs to say the second sentence, not the first.
Where we stand
The finished phone AI has now sat parked eighteen days Clock running
The phone AI
Built, tested, answering a real number. Eighteenth straight day at the top of the board, unshipped.
Receptionist positioning
Moved to #1 on the action list the day Wix surfaced. The competitor is no longer four startups — it is a feature the prospect may already pay for.
Image side-by-side
#2 — run Squarespace's Image Studio against one of our venue photo sets and see honestly who wins.
Governance one-pager
Fourteen days owed, and as of this week nothing is blocking it.
Venue-trained receptionist
The version Wix cannot copy is unbuilt for the twentieth week.
Bottom line: last week we said a quiet week was the time to clear old work. This week was not quiet, and the old work got more urgent, not less.
Anthropic shipped Claude Opus 5.5 on Sept 22. Anthropic's claim: the quality of its top model at about 40% less cost on typical work.
The line that matters to us fell 60%. Our site rebuilds and research runs are mostly re-reading the same long context — that is the price that dropped most.
Same budget, about 1.7 times the refinement passes per site. Quality is how many passes we can afford, so this is a quality upgrade we pay less for.
New control switches landed Sept 25–26 that let us lock which AI models a client's tools may use. That unblocks the governance offering on card 6.
Bottom line: cheaper and better on the same day. The only way to waste it is to keep running last month's settings — switch the default this week.
Platform · Cloudflare
Vantage is fine — and Cloudflare gave us the safe test space we never had All calm
Nothing on our platform broke or is scheduled to break. Three notes.
The good one
Worker Previews (Sept 22): every code branch gets its own isolated copy of the platform, with its own data. A change that could overwrite an owner's edits now fails in a sandbox instead of on Magnolia House. It adds a rehearsal step; our deploy discipline stays as is.
A correction
We warned for several days that Tuesday's security-filter update could block client pages. Checked against the source, it is almost entirely watch-only. One small rule merge is worth a glance; nothing more.
Long view
Cloudflare expects automated AI traffic to hit a thousand times human traffic within five years. Someday, how a venue site treats AI crawlers becomes a revenue setting. No action this quarter.
Bottom line: the platform got safer to change. Use Previews on the next tenant update and the "never blind reseed" rule gets a real safety net.
Biggest growth move · Managed IT
The AI-agent governance offer is ready to write — controls and the reason, both in hand Act now
For two weeks our one-page "governing AI agents" offer for Managed IT clients was waiting on two things. Both arrived.
The controls: the new switches from card 4 — lock which models are allowed, log what the agents do.
The reason: Anthropic disclosed a fourth real incident where an early model, thinking it was in a test, got admin access to a real company's system and changed settings. No malice — just too much access and a wrong assumption. That is the sentence a business owner understands.
The market backs the move. A survey of 600-plus top-ranked MSPs: recurring revenue grew 11.9%, faster than total revenue at 9.6%; 91% now offer AI; 57% expect AI to be among their biggest revenue gainers next year, second only to managed security. It is a self-selected list of top performers, so treat it as the ceiling rather than the average.
Bottom line: recurring security-style work is the revenue that earns a multiple. This one-pager turns AI worry into a monthly line item — write it this week.
Client-facing · venues & Vantage
Wix's phone agent only knows the website. Ours has to know the venue. Reposition
Wix's new phone agent answers from the site's published pages only. It does not know the venue's past leads, which tours booked, which dates are soft, or what the owner will flex on.
Our pitch changes today. "An AI answers your phone" is no longer something a venue will pay us for. "An AI that knows your bookings and your history" still is.
A plan correction. The low-cost model we planned to test in the voice system is not offered there. We test the two that are, and score them on questions only venue memory can answer, not on cost per call.
Imagery: we run Squarespace's Image Studio against a real venue photo set, side by side with our work, before a prospect does it for us.
Bottom line: the four-second test a venue owner runs — does it know my business? — is now the only test that matters. Build and demo to that.
The wider AI race · so what
A same-day price war means generic AI keeps getting cheaper — and less of a moat Context
On Sept 22, Anthropic and OpenAI both cut prices. OpenAI's new budget model costs about a fortieth of Opus 5.5 per word; its mid-tier costs half.
OpenAI's benchmark claims are aggressive and self-reported. We don't take either vendor's scoreboard at face value — our own test runs both on the same rebuild job with the same scorecard.
The so-what: when intelligence costs almost nothing, builders bundle it for free — which is exactly what cards 2 and 7 showed. The value moves to whose data it runs on and who is accountable when it's wrong.
On the ownership side: a founder-led buyer, not private equity, has done 48 acquisitions in under four years, targeting firms around one to five million in revenue. Useful context for our own acquisition thinking.
Bottom line: falling prices help our costs and hurt anyone selling generic AI. Make sure we are never the one selling generic AI.
The bottom line
Two decisions need Bob and Niven this week Decisions
1 · Set a ship date
The phone AI has been finished and parked eighteen days. Every week it sits, the generic version gets given away by someone else. It needs a date on a calendar, not more build.
2 · Approve the new positioning
Stop pricing against "an AI answers your phone." Sell the receptionist that knows the venue's leads, calls and tours — and agree that building that memory is the next build. Twenty weeks unbuilt.
What Paul does without you this week: switch our tools to the cheaper model, write the governance one-pager, run the image side-by-side, and use Worker Previews on the next tenant change.
Bottom line: the market moved three times in one week. Two yes-or-no answers from you keep us ahead of it instead of behind it.
Card 1 / 9 · ~20s
Narration audio not found yet — run the render step, then reopen.