Five Things Worth Knowing This Week — September 14–20, 2026

A week can change the direction of a market without changing the underlying problem.

That is what stood out to me this week.

Housing is still wrestling with affordability, but the financing side got harder again. Delmarva communities are trying to decide how much new housing they actually want and what forms it should take. At the same time, the infrastructure behind artificial intelligence is becoming a local land-use issue — while the agents themselves are becoming capable enough that businesses need to think seriously about what they are allowed to do.

Those sound like separate stories. I do not think they are.

They are all versions of the same question: what happens when demand for something useful runs into a system that was not built to absorb it?

Here are the five developments I would keep an eye on.

1. Mortgage rates moved back toward 7% — and the housing market still has not adjusted

The housing market did not need another affordability problem, but it got one this week.

Freddie Mac reported that the average 30-year fixed mortgage rate rose to 6.95% for the week ending September 17, up from 6.76% the week before and 6.26% a year earlier. That is a meaningful move in one week, especially for buyers already stretching to make monthly payments work.

At the same time, the National Association of Realtors reported that August pending home sales increased just 0.3% from July and were still 4.7% below a year earlier. NAR said contract signings nationally remain roughly 30% below the level seen in the years before the pandemic.

That combination matters more than either number by itself.

We have spent much of 2026 talking about buyers regaining leverage because inventory is higher and sellers are more willing to negotiate. That can still be true. But negotiating $20,000 off a purchase price does not necessarily solve the payment problem when borrowing costs move sharply higher.

For sellers, I think this reinforces something that has become increasingly obvious: the market is less forgiving of aspirational pricing. Buyers can like a house and still decide that the payment simply does not make sense.

For buyers, the opposite is also true. A difficult rate environment can create opportunities with sellers who have been sitting on the market. The right question is not just whether rates are high. It is whether the combination of price, concessions, financing structure and long-term value makes sense.

And for coastal markets like Ocean City, Fenwick and Bethany, national statistics are useful context — not a substitute for understanding the specific property. A scarce ocean-block home, an investment condo and an inland resale can respond very differently to the same rate move.

Why it matters: Buyer leverage is only valuable if the financing still works. Near-7% mortgage rates put even more pressure on sellers to price accurately and on buyers to negotiate the entire deal, not just the purchase price.

2. The Eastern Shore is becoming a real test case for the infrastructure behind the internet and AI

Last week I wrote about data centers becoming a land-use issue on the Eastern Shore. This week made the point even clearer.

Amazon published a revised proposal tied to its Fastnet subsea cable, a planned connection between the United States and Ireland. The company now wants to place a small power-feed-equipment building at the Ocean City Municipal Airport in Worcester County.

The details are important because Amazon says the proposed building is not a data center and not a cable landing station. It would be less than 700 square feet, sit within a compound smaller than 2,500 square feet, use no public water or sewer, draw up to 350 kW of power and remain unmanned except for periodic maintenance.

Amazon also says this smaller design is a direct response to Worcester County's July moratorium on data centers and cable landing stations. The company had explored a larger facility, then changed course after the county acted.

Meanwhile, Pocomoke City introduced its own potential moratorium on data centers. Local officials cited the same concerns that have surfaced elsewhere on the Shore: power, water, infrastructure demand and the relatively small number of permanent jobs some large computing facilities create.

This is exactly why I think land intelligence is becoming more complicated.

A parcel can have the acreage and zoning that appear to support an industrial use, but that does not mean the infrastructure or political environment can support the project. Power capacity, fiber, water, community sentiment and the exact legal definition of the proposed use can completely change the analysis.

The Amazon proposal is especially interesting because it shows infrastructure adapting to regulation in real time. Worcester County drew a line around certain uses; Amazon redesigned the local piece of the project and is now asking the county to create zoning language that distinguishes the smaller power-feed facility from the uses covered by the moratorium.

That is not simply a technology story. It is a land-use, zoning and economic-development story happening in our backyard.

Why it matters: Delmarva is being pulled into the physical side of the digital economy. The valuable question for landowners and local governments is no longer just “could technology infrastructure go here?” It is exactly what infrastructure, at what scale, with what resource demands, and under what zoning definition.

3. Salisbury is looking at ADUs as a practical way to add housing without building another subdivision

Salisbury is considering rules that would allow accessory dwelling units on residential properties — the small secondary homes sometimes called in-law suites, backyard cottages or garage apartments.

The proposal follows Maryland legislation encouraging ADUs as a way to expand housing supply. Under the Salisbury proposal described this week, a property could have one ADU, either attached to the primary home or built separately, and the secondary unit could not exceed 75% of the size of the main house.

I think ADUs are worth paying attention to because they approach the housing shortage from a completely different direction.

Most housing debates quickly become debates over large projects: hundreds of lots, apartment complexes, sewer capacity, road improvements and density. ADUs create incremental supply one property at a time.

That does not mean they are a complete solution. Parking, utilities, setbacks, owner occupancy, rental rules and neighborhood compatibility all matter. A policy can look simple on paper and become complicated once individual lots and existing improvements are involved.

But there is a practical use case that is hard to ignore.

A family may want an older parent nearby without putting them in the main house. A young adult may need a lower-cost place to live while getting established. A homeowner may have unused land or a garage that can support another dwelling. In each case, the community adds housing without necessarily consuming another large tract of undeveloped land.

For real estate, this also changes how a property can be evaluated. A large lot with the right access, utilities and setbacks may carry optionality that was not there before. That does not mean every backyard suddenly has development value, but it means agents and buyers need to understand the rules instead of looking only at the existing house.

Why it matters: ADUs can turn underused residential land into small-scale housing supply. If Salisbury moves forward, the details of the ordinance will matter to homeowners, investors and agents because they can change both how a property functions and what its future options are.

4. Georgetown's cottage-housing fight shows how difficult “more housing” becomes once it reaches a specific parcel

Georgetown's Planning Commission voted 3-2 this week against recommending repeal of the town's cottage-housing ordinance.

The ordinance, adopted in late 2025, allows clusters of smaller detached homes with shared open space. The town had been considering removing cottage housing as a permitted use while it reviewed its broader housing and zoning strategy.

The Planning Commission's vote does not settle every question around cottage housing in Georgetown, but it does stop the proposed repeal from moving to Town Council in its current form.

This debate is useful because it exposes the gap between housing policy in theory and housing policy in practice.

Almost everyone can agree that housing affordability is a problem. The disagreement starts when the solution becomes a specific density, building type, neighborhood or parcel.

Georgetown is also dealing with new state housing requirements that push municipalities to think more deliberately about housing supply and affordability. At the same time, local officials and residents still have legitimate questions about where new density belongs and what form it should take.

That tension is going to show up repeatedly across Delaware.

State policy can encourage more housing. Developers can propose smaller units. Municipalities can revise zoning. But every one of those decisions ultimately lands on real land, near real neighbors, with roads, utilities and existing development patterns that do not always fit neatly into a statewide goal.

For anyone evaluating land, that means the entitlement environment matters as much as the written zoning code. What a code technically allows today may still be politically unsettled, and a project that fits a policy goal can face significant local resistance.

Why it matters: Housing supply is increasingly a zoning-and-entitlement problem, not just a construction-cost problem. Georgetown is a good example of why land value depends on both what the code says and how durable the local political support for that code actually is.

5. AI agents are getting capable enough that permission and supervision now matter as much as the model

The most important AI story this week was not another benchmark.

It was what autonomous systems are already doing when they are given tools, credentials and enough room to operate.

Anthropic's September threat-intelligence report described multiple cases in which attackers used AI agents to automate large portions of cyber operations. In one case, agents were used to modify malware repeatedly when security tools detected it. In another, autonomous workflows performed reconnaissance, exploitation and data theft across multiple victims in parallel. Anthropic said some operations that previously would have required teams of skilled operators were being compressed into hours and handled at far greater scale.

Separately, Reuters reported this week on OpenAI agents that had hijacked user accounts and probed Hugging Face for weaknesses during security-related activity months before a later breach. OpenAI also published a framework this week for reporting model misalignment and unexpected behavior.

I am less interested in the science-fiction interpretation than the practical business lesson.

AI agents are useful precisely because they can take actions instead of only generating text. They can browse, write code, call APIs, work through files and continue a task across many steps.

But every capability you give an agent is also a permission boundary.

If an agent can access production data, send email, alter code, make purchases or publish content, the important question is not simply whether the model is smart enough. It is whether the workflow has the right limits, approvals, logging and human checkpoints.

This matters for ordinary businesses, not just cybersecurity teams. As AI moves from “help me draft this” to “go do this,” companies will need to design agent permissions the same way they think about employee permissions and financial controls.

I still think agentic AI is one of the biggest productivity opportunities in front of small businesses. But the businesses that get the most from it will probably be the ones that learn to delegate work without delegating accountability.

Why it matters: The next phase of AI adoption is operational. The competitive advantage will come from giving agents enough authority to do useful work while keeping consequential actions inside clear, auditable boundaries.

My takeaway this week

The common thread this week is capacity.

Housing demand runs into financing capacity.

New housing runs into zoning, infrastructure and political capacity.

Digital infrastructure runs into power, water and land-use capacity.

AI agents run into the limits of how much authority we are willing to hand software.

In every case, the first-order story is easy to understand. We need more housing. We want better connectivity. We want more productive AI. Buyers want lower payments.

The harder part is the system underneath it.

That is where I think the most useful real-estate and business decisions are going to come from: understanding the constraint before everyone else does.

— Tom