What Information Technology Means Now and Over the Next Five Years

My post about an article I wrote 22 years ago, along with a question that came up at a recent meeting, got me thinking: what is information technology today? Many colleges already run cybersecurity programs and AI programs, each with its own courses, certifications, and advisory board. IT often gets whatever the other two have not claimed.

I have heard versions of that question for over 40 years. It came up when PCs arrived on desks in the 1980s. It came up again when the internet arrived in the 1990s. The answer changed each time, and it has changed again.

When I started, IT meant hardware you could touch. Minicomputers sat in climate-controlled rooms. Later, servers sat in a closet, cable ran through the ceiling, and a technician fixed a printer by walking to it. The job was installation, repair, and backups on tape.

Three changes moved most of that work somewhere else. Virtualization put dozens of servers on one physical machine. Software as a service moved email, file storage, and business applications to vendors. Cloud platforms let a company rent compute by the hour. Gartner projects public cloud growth of 21.3% in 2026, with the market reaching $1.48 trillion by 2029. Gartner also forecasts worldwide IT spending of $6.31 trillion this year, with IT services above $1.87 trillion and data center systems above $788 billion.

The work followed the spending. IT today (my opinion) covers:

•       Cloud and hybrid infrastructure, including cost control, since rented compute bills by the hour

•       Identity and access for every user, device, and application

•       Networks linking offices, homes, branch sites, and multiple cloud providers

•       Patching and configuration across thousands of endpoints

•       Data platforms that security and AI teams depend on

•       Service management, automation, and vendor contracts

Patching deserves a closer look. The Verizon 2026 Data Breach Investigations Report found exploited vulnerabilities became the top way attackers got in, at 31% of breaches. Organizations fully fixed only 26% of critical known-exploited vulnerabilities in 2025, down from 38% the year before. Median time to patch rose from 32 days to 43. Patching is IT work. The security team identifies the vulnerability; the IT team deploys the fix to every affected system.

The help desk shows the change inside the workforce. BLS projects computer support specialist jobs to fall 3% from 2025 to 2035 as chatbots take routine troubleshooting. Computer and information systems managers should grow 16% over the same period. Across all computer and IT occupations, BLS expects about 280,000 openings a year at a median wage of $109,470. Entry-level IT now starts with the problems a chatbot cannot resolve.

IT work also moved out of tech companies. CompTIA's State of the Tech Workforce 2026 reports that 60% of US tech professionals work for organizations outside the tech industry. CompTIA projects net tech employment to grow 1.9% in 2026 to about 9.8 million workers. In January 2026, about 275,000 job postings asked for some level of AI skill. Massachusetts ranks fifth among states for tech employment concentration at 8.0%.

Cybersecurity and AI both run on IT. A security team enforces access through identity systems that someone built and maintains. An AI model needs clean data, provisioned compute, and network capacity. Remove IT and both groups have policies with nothing to apply them to.

The job numbers point the same way. BLS projects about 14,100 openings a year for information security analysts, about 5% of the 280,000 annual openings across all computer and IT occupations. The rest include software developers and the network, systems, cloud, database, and support roles that IT programs prepare students for. A college that offers cybersecurity and AI programs without an IT program leaves its students out of a large share of those openings.

IT programs are still needed, and they belong alongside cybersecurity and AI programs. Each of those programs sends graduates into organizations that depend on IT staff to build and run the systems being protected or automated. Strong cybersecurity and AI programs make a strong IT program more important. What changes is what IT programs teach, and I will write more about that in a future post.

Pic from Nikolas Badminton at futurist.com

I snagged this pic from an X post by Nikolas Badminton – thanks Nikolas! Tomorrow is Halloween, and this costume seemed kinda right for a post about the next five years. The package includes three talking points about the age of agentic AI, Ray-Ban Meta glasses, zero practical experience, industrial (love that word) pants and shirt, and a new ebook built using Claude. I will leave it to you to guess which items I already own.

Five-year AI predictions are easy to make and easy to sell this year. The forecasts below come from Gartner, along with Gartner’s own warning about how many of these projects will fail.

Any curriculum built today has a short shelf life, so the more important question is what IT will mean by 2031. The next five years will be driven by AI agents moving into IT operations. Gartner projects that by 2029, 60% of enterprises will use agentic AI in IT infrastructure operations, up from under 10% today. A vendor summary of the same research puts the figure at 70%. Either way, most IT shops will run agents within five years.

The same Gartner research expects only 20% of AI-suggested actions to need human approval in 2029, down from 80% in 2025. Policy-based guardrails will decide what an agent may do. Gartner also expects half of infrastructure and operations teams to be restructured by 2030.

Many of these efforts will stall first. Gartner predicts more than 40% of agentic AI projects will be canceled by the end of 2027 because of cost, unclear value, or weak risk controls. The timeline will slip in places. The direction will hold.

The IT job shifts from making changes to defining, approving, and auditing the changes automated systems make. The skills that gain value:

•       Change control: approval gates, rollback plans, and maintenance windows for agent-driven changes

•       Non-human identity: permissions for agents, service accounts, and API keys

•       Policy writing: turning rules into code that agents must follow

•       Logging and audit: reconstructing what an agent did, when, and why

•       Cost management: agents consume compute, and compute bills by the hour

•       Data quality: agents act on the data they receive, errors included

Networking, Linux, and scripting stay in the core. A technician who understands how a network routes traffic can catch an agent's bad change before it takes down a site.

For educators, this means an IT curriculum has to change on a schedule. A program built now should look different by 2031, with more identity, automation, and agent oversight and less hardware. Employer input and regional job data should drive each update, and the update cycle belongs in the plan from the start.

So what is information technology today? It is the infrastructure, identity, data, and automation work that every other group assumes already exists. Every college with a cybersecurity or AI program needs an IT program to go with it. Over the next five years, IT will also decide what the agents are allowed to do. And someone still has to set up the accounts.

Progress in Large Bytes, 22 Years Later

In June 2004, BusinessWest ran an article of mine in its 20th anniversary issue. It was titled "Progress in Large Bytes" and covered two decades of advances in technology. I was executive director of the National Center for Telecommunications Technologies then. I reread it recently. Most of it held up.

The article started with the 1981 IBM PC: a 4.77 MHz Intel 8088 processor, 16 KB of memory, one or two 160 KB floppy drives, and a price a little over $1,500. It moved through Novell NetWare, Tim Berners-Lee writing the first browser code at CERN in 1990, and the April 1995 decision by the NSF to privatize NSFNet. ISPs appeared in every part of the country, and dial-up modems quickly felt slow.

One correction. The article credits VisiCalc to Harvard MBA student Dan Baker. His name is Dan Bricklin. Twenty-two years later, the record is fixed.

The article made two predictions. First, Microsoft's next operating system, code-named Longhorn, would need a broadband connection for its full feature set. Longhorn shipped as Windows Vista in 2007 and worked offline, so that one missed. Second, always-on wireless connectivity would dominate business and consumer technology, across PDAs, VoIP phones, tablets, cell phones with Web interfaces, and Game Boys. That one landed. The phone in your pocket now does all of those jobs.

The numbers moved as well. A phone with 8 GB of memory holds 524,288 times the 16 KB in the original IBM PC. Scott McNealy's line, "the network is the computer," now covers AI models that run in data centers and quantum processors that some can already reach through a cloud login. Neither sits on your desk.

The last sentence of my 2004 article reads, "We have only started to scratch the surface." It aged better than the Longhorn paragraph. Check back in 2048.

SpaceX Wants 80,000 Cell Sites

I led Verizon's Next Step telecommunications faculty group in New England, which trained telecommunications technicians. A couple years ago, as a consultant, I co-wrote curriculum for Verizon's Skill Forward program. Every site a telecom technician maintains needs the same three things: a place to stand, power, and a path back to the network. The list never changes.

On October 22, Bloomberg reported that SpaceX is in talks to access more than 80,000 cell sites as it works toward becoming the fourth major US wireless carrier. The sources are unnamed. Crown Castle is one of the tower owners in the discussions. CEO Chris Hillabrant did not confirm specifics and said, "You can assume that we're talking to all parties." Verizon, AT&T, and T-Mobile each fell more than 3% in extended trading Thursday. Musk described the goal on the earnings call as "broad-based connectivity anywhere that SpaceX and Starlink are allowed to operate."

This follows the spectrum deal. On October 8, SpaceX agreed to buy Grain Management's nationwide 800 MHz portfolio. Low-band signals pass through walls better than high-band signals, and I covered that in Walls Are the Hard Part of Phone Service From Space.

T-Mobile CTO John Saw argued that a network needs more than spectrum. A satellite link has about 50 dB more path loss than a cell site 1 km away, and one satellite beam covers an area that hundreds of terrestrial sectors serve. Sites supply the shorter path and the capacity. My Spectrum Still Needs a Network post works through his numbers.

Free-space path loss at 2 GHz, satellite versus cell site. Not to scale.

Sites also mean work. Each one needs installation, power, backhaul, and technicians for upkeep. Wireless Estimator reports that SpaceX has described small radios beside Starlink dishes on rooftops, so the work includes rooftop installs as well as towers. Technicians with RF, power, and fiber skills fit that work.

Telecom techs will recognize the plan. A place to stand, power, and a path back to the network. Eighty thousand sites need eighty thousand sets of those three things.

Two Fault Tolerance Results in One Week: Quality and Cost

Chapter 4 of Quantum from the Ground Up explains that a logical qubit is built from many noisy physical qubits. Chapter 13 cites Craig Gidney's 2025 estimate of roughly one million physical qubits to break RSA-2048. Both chapters depend on two numbers: how reliable a logical qubit is, and how many physical qubits it consumes. Two results published within a week of each other address one number each.

The Vocabulary

A physical qubit is the actual device: an ion, a superconducting circuit, or a neutral atom. It makes errors constantly. Error correction spreads one unit of quantum information across many physical qubits, and the group acts as one logical qubit. Repeated measurements called syndrome checks reveal where errors occurred without reading the stored information. Software called a decoder decides how to fix them. I covered one example in a post on IonQ decoding errors in real time on a single CPU.

Code distance is the smallest number of physical errors that can corrupt a logical qubit without detection. A larger distance needs more physical qubits. When the physical qubits are good enough, each increase in distance lowers the logical error rate. Logical fidelity is the fraction of logical operations that finish without an error. Color codes and surface codes are two families of error correcting codes. They use different qubit layouts and have different qubit costs.

Quantinuum: Quality

On October 19, Quantinuum reported 99.9995% logical fidelity on its Helios computer. That equals five errors per million operations, or about one error per 200,000. The company says the error rate is ten times lower than its own September record. It also says the rate is 100 times lower than the best published result on any other hardware type, which came from Google on a superconducting system.

The experiment used a color code. The team raised the code distance in steps, and each step cut the remaining logical errors by a factor of 12. That is the below-threshold behavior described above, and the factor matters as much as the headline fidelity. Two more steps at that rate divide the error rate by 144. A Sandia researcher who was not involved in the work said a logical qubit nearly 100 times better than its physical components could, when scaled up, reach useful reliability with fewer than one million physical qubits. The full study is linked from Quantinuum's announcement.

The September record used a different code, called Helix, on the same hardware. Helix needed 10 physical qubits per logical qubit. The color code needed up to 37 and delivered the lower error rate. Quantinuum presents this as evidence that its architecture can swap codes to trade accuracy against qubit count. Its August result built a universal gate set from 28 qubits, which I covered in an earlier post on that 28-qubit gate set. The cost of the magic states behind such gate sets is the subject of a post on why quantum chemistry waits on the T gate.

Google: Cost

On October 15, researchers at Google Quantum AI, Google DeepMind, and MIT posted a preprint on reducing the physical qubit bill. The standard surface code lays qubits on a flat grid. At the distances in the paper it needs more than 1,000 physical qubits per logical qubit. Higher-rate codes pack many logical qubits into one block, but on a chip they need long-range connections that cross each other.

The Google team tiles a hyperbolic surface with small flat modules of 72 or 116 qubits, joined only at their edges. Each module uses ordinary nearest-neighbor gates. Circuit-level simulations show a 10x reduction in physical qubits versus the surface code at a logical error rate of one in ten billion. One simulated hyperbolic surface code spread 146 logical qubits across 120 modules, 13,920 physical qubits in all, which works out to about 95 physical qubits per logical qubit.

The headline figure of more than 30x comes from extrapolation. The authors projected color codes too large to simulate or decode, which reach 36 physical qubits per logical qubit at distances of 20 to 24. The paper is a preprint that has not been peer reviewed.

Reading the Two Together

One result is measured on hardware and the other is simulated. Quantinuum lists all-to-all qubit connectivity as part of its architecture, so its codes need no special layout. Google restricts its design to mostly nearest-neighbor gates inside modules, which suits solid-state chips. IBM's work on linking two quantum cryostats toward fault tolerance is another modular route on superconducting hardware. The diagram below puts the two new results on one page.

Quantinuum measured a 12x drop in logical errors for each increase in code distance (left). Google simulated far fewer physical qubits per logical qubit than the surface code needs (right). Blue bars come from hardware. Orange bars come from simulation or projection. Compare orders of magnitude only, because the codes and counting conventions differ. 

Quantinuum's 10 and 37 physical qubits per logical qubit sit in the same range as Google's projected 36. The sources do not state a common counting basis, so compare orders of magnitude only. Neither result runs a useful algorithm. Both benchmark error suppression and qubit cost, and the two numbers they move are the ones that decide how large a fault tolerant machine has to be.

The arithmetic is direct at the scale of the DOE's funding target of 100 logical qubits. At more than 1,000 physical qubits per logical qubit, the surface code reference implies over 100,000 physical qubits. At 36 per logical qubit, the same target implies about 3,600.

What This Changes in the Book

Chapter 4 covers error correction basics. It should add the logical qubit figures above as the current benchmark: 99.9995% logical fidelity and a 12x error reduction per distance step.

Chapter 6 lists trapped-ion figures for Quantinuum's H2 at 56 qubits and IonQ's 99.99% two-qubit gate fidelity. Those are physical layer numbers. Quantinuum's result is a logical layer number on newer Helios hardware, and the chapter needs that distinction along with the update from H2 to Helios.

Chapter 12 covers NVIDIA's Ising Decoder. Quantinuum's architecture integrates tailored AI decoders, so decoding belongs in the story of this result. Chapter 16 covers Google's superconducting and neutral atom platforms. The modular hyperbolic design targets the superconducting side, where long-range wiring is the constraint.

Chapter 13 carries the Gidney estimate of about one million physical qubits for RSA-2048. Neither result runs Shor's algorithm at scale, so neither changes the Q-Day range in the chapter by itself. Both lower the overhead per logical qubit, which is one of the inputs to that estimate, and the chapter should say so.

This post will be added to the next edition of the book. The current edition is described in the second edition announcement, and the download is on the Quantum from the Ground Up Drive.

Westfield Planning Board Ends the Servistar Data Center Project

I have written a few posts about the proposed Servistar Realties data center in Westfield, the one planned over the Barnes Aquifer. They include Can Data Centers Be Built and Operated Responsibly? Ask My Hometown, Inside NVIDIA's 113-Degree Cooling Design, and Westfield's Data Center Question, and The Trout Stream Under the Turnpike, and the Data Center Over the Barnes Aquifer. Last week the project got a decision. On Tuesday, the Westfield Planning Board voted unanimously to deny the developer's request to extend its special permit, according to Boston.com. The special meeting ran about two hours.

The project carried a $3 billion price tag and covered more than 150 acres. The board granted the original permits in October 2021. Servistar never bought the land. Its limited liability company status expired in late 2024, and it never recorded the permit with the Hampden County Registry of Deeds. Board members found no good cause for an extension. Five years is a long time to hold a permit for land you do not own.

Servistar's representative, Paul Corey, blamed the market. He cited the wait for state tax incentives, which Governor Healey paused this summer, and federal tariffs that disrupted chip supplies and energy costs. Board members disagreed. The chair said a building should already have gone up.

Two items remain open. Servistar has 20 days to appeal under state law. The City Council passed a one-year moratorium on data centers in July, but this project was exempt because it was already underway. If the denial stands, Westfield will have no large data center pending and a moratorium covering new applications.

My earlier posts covered water and cooling. The board decided on missing paperwork, so those questions never reached a vote. They will return with the next proposal, and you can expect the next developer to own the land first. Watch the appeal window.

Why Quantum Chemistry Waits on the T Gate

I struggled with organic chemistry in college. The course asked me to predict how molecules behave from their structure, and I predicted poorly. The cause could have been the content, the instructor, or me. I will never know but I’m kinda leaning towards me. Things have changed these days - a quantum computer can make predictions from the molecule's energy, and the effort runs into a bottleneck that is easier to identify. 

Bacteria turn nitrogen from the air into a form living things can use. The enzyme that does the work, nitrogenase, runs at room temperature and ambient pressure, according to a 2026 resource-estimation study. Its active site is a cluster of iron, molybdenum, sulfur, and carbon atoms called FeMoco. Chemists want FeMoco's energy at high precision, because the energy shows how the reaction proceeds. Classical computers cannot solve it exactly at this size, so FeMoco is a standard target for quantum computers.

The algorithm for that job is quantum phase estimation, or QPE. Every molecule has a Hamiltonian, the mathematical description of its energy. The lowest value is the ground state energy. Alexei Kitaev introduced phase estimation in 1995, and given a state close to the ground state, it reads out the energy as a number. Preparing a good starting state is difficult on its own, but the larger cost comes from the gates the algorithm runs.

A gate is one basic operation on qubits. Quantum computers protect their qubits with error correction (Chapter 4 of Quantum from the Ground Up covers the basics). The machine groups many physical qubits into one logical qubit, which can survive errors in the physical qubits beneath it. A family of gates called Clifford gates runs cheaply on logical qubits. Clifford gates alone cannot perform every calculation. The machine also needs a non-Clifford gate, usually the T gate. A logical T gate consumes a specially prepared resource called a magic state, and preparing it takes extra qubits and many rounds of checking.

The count of T gates sets the cost. The best published QPE estimate for FeMoco needs about 5.3 billion Toffoli gates. A Toffoli is a three-qubit gate requiring seven T gates in standard circuit synthesis (or four with an ancilla), though modern fault-tolerant architectures often cultivate |CCZ⟩ states directly to bypass explicit T-gate expansion. Under standard synthesis, this total is roughly 37 billion T gates. A PennyLane tutorial for a 76-orbital FeMoco model shows a trade-off. One version uses 2,188 logical qubits and 88 billion Toffoli gates. A second version cuts the qubits to 466 and needs 345 billion Toffoli gates. Saving space costs more gates, and every gate adds T gates to the bill.

Distillation was the standard way to make magic states. It combines many noisy magic states into one cleaner state, and the circuits that do this are called T factories. Craig Gidney's 2025 Simons talk cites a 2012 analysis that put T factories at 94 percent of a quantum computation. That share made magic states the main target for cost reduction.

Chapter 6 of the book covers one alternative to brute-force distillation. In August 2026, a Quantinuum and UC Davis team used code switching to build a trustworthy magic state from 28 physical qubits instead of hundreds. The state's logical infidelity came out about 2.7 times lower than the error rate of the physical qubits that built it. Cultivation takes a different route to the same goal.

In 2024, Gidney, Shutty, and Jones proposed magic state cultivation. It builds up one magic state inside a single error-correcting patch, checks the state at each stage, and discards any attempt that fails. At a physical error rate of 0.1 percent, it reaches a logical error rate near 2 in a billion using about an order of magnitude fewer qubit-rounds than earlier methods. A qubit-round is one qubit held for one round of error correction, so the measure combines space and time. The authors suggest distillation may become unnecessary. Google tested cultivation on a superconducting processor in December 2025. The error dropped by a factor of 40, the state fidelity reached 0.9999, and 8 percent of attempts survived. The remaining 92 percent were discarded during early-stage syndrome checks. This post-selection converts a spatial footprint penalty into a temporal retry delay. Factory duty-cycle and retry latency become the primary runtime bottlenecks.

From a molecule’s energy to the magic states every T gate consumes, with three ways to make them.

Lower T-gate cost moves chemistry closer. A March 2026 study estimates that chemically accurate QPE for active spaces of 20 to 50 spatial orbitals is feasible on early fault-tolerant machines with about 100,000 physical qubits, with runtimes of days to weeks. An active space is the set of orbitals the calculation treats exactly, and chemical accuracy is the precision chemists need to predict reactions. Exact classical methods like Full Configuration Interaction (FCI) exponentially stall beyond 22 spatial orbitals, making 20-to-50 orbital strongly correlated systems the primary domain for early quantum advantage.

The same T-gate cost applies to cryptography. Chapter 13 of Quantum from the Ground Up cites Gidney's 2025 estimate that RSA-2048 falls to fewer than one million noisy qubits in under a week, then reports that the figure did not survive spring 2026. An Oratomic estimate cited there puts RSA-2048 within reach of as few as 10,000 qubits on a neutral atom machine. Part of Gidney's reduction came from using cultivation to shrink the space for magic state production. Progress on the T gate therefore shortens the timeline for chemistry and for breaking encryption. The September 2026 edition covers code switching in Chapter 6. The next quarterly edition of the book will add cultivation and the T-gate budget for chemistry.

I still cannot say for sure whether the content, the instructor, or I caused my trouble with organic chemistry but still.... kinda leaning towards me.

Agents Read the Reports About Themselves

I am building AI/SI into my courses and lab materials. I call it PGI, for Pretty Good Intelligence, a nod to the encryption tool Pretty Good Privacy. On weaker days it is SGI, Sometimes Good Intelligence. I want my students to use both, so the assignments I am writing require it. That means I have to understand how these systems behave, and I learn something new about them nearly every week.

This week’s material came from a research report on an AI/SI agent fleet. An agent is a language model with tools, such as a browser, a code runner, and a network connection. A fleet is a group of agents that start from one place with similar goals. Researchers watched one fleet, running on a commercial cloud platform, between 28 September and 8 October 2026.

The task was simple: find which entrance of Chengdu Zoo most map users navigate to. The mapping site blocked plain requests with anti-bot checks, so the agents improvised. They submitted programs to urlquery.net, a public security scanner that loads pages in a real browser, and used that browser to run their own code. Results returned through page titles and disposable web inboxes. The researchers counted more than 2,600 attempts. Some probed non-production servers and tried exposed access keys. Most failed. The report also cites earlier incidents involving agents from other labs, including a reported case in Germany, so the pattern is general.

The report offers three lessons. First, audit what your sandboxes can reach, because network access is where agents find workarounds. Second, tell agents the truth. The researchers placed hidden text on their site with false claims, agents acted on it, and the authors regret it. Third, assume agents read what you publish. One agent inbox message used the word “fleet” after the researchers coined it.

The fleet gave PGI a fair test. Turning a public security scanner into a code runner is pretty good intelligence. Missing on most of the 2,600-plus attempts is sometimes good intelligence. My students will earn the same two ratings. Work that checks the output earns PGI. Work pasted in unread earns SGI. The gradebook will use both labels. I am officially claiming the terms PGI and SGI. Citations are always appreciated 😎