Chemists

Three parts of your work where AI is already doing real lifting, and what stays yours.

• Fire1720–1820

  Fire, balance, and glass: the classical laboratory (furnace-centered design)

  The 18th-century chemist worked in a furnace-centered laboratory descended from alchemical workshops: ceramic crucibles, retorts, and alembics for distillation; balance and weights for quantitative measurement; simple acids (sulfuric, nitric, hydrochloric) produced from vitriol and saltpeter. The Mettler analytical balance, achieving milligram precision, and Scheele's isolation of chlorine, manganese, and tartaric acid demonstrated what careful wet chemistry could do without any instrument beyond glass and fire. Lavoisier's greatest innovation was methodological: he demanded precise before-and-after mass measurements for every reaction, turning chemistry from a qualitative art into a quantitative science. No significant labor displacement occurred in this era -- the chemist's work was inherently artisanal and irreplaceable.

• Bunsen burner bench laboratory1820–1950

  Bunsen burner bench laboratory and spectroscopic identification (Liebig-era design)

  Robert Bunsen and Gustav Kirchhoff developed emission spectroscopy in 1859 at Heidelberg, showing each element emits a unique spectral fingerprint. Coupled with the gas burner that Bunsen and Peter Desaga perfected (producing a controllable, soot-free flame), this gave chemists a fundamentally new identification tool that could detect trace elements in complex mixtures far faster than wet gravimetric methods. Justus von Liebig's Giessen laboratory, opened around 1826, established the pedagogical model of bench-based systematic synthesis and analysis that spread to every major university by 1870. By 1870 the laboratory design had stabilized: central bench, fume hood, reagent bottles, gas line, and a small spectroscope. This infrastructure remained essentially standard until the mid-20th century. The spectroscope sharply reduced the time required for elemental identification and enabled new synthetic programs (discovering caesium and rubidium in 1860-61, for example), but it augmented rather than displaced the trained chemist.

• Instrumental analysis revolution1950–1985

  Instrumental analysis revolution: IR, UV-Vis, GC, NMR, and mass spectrometry

  Between 1950 and 1970, a cascade of new instruments transformed analytical chemistry almost beyond recognition. Gas chromatography (invented by Martin and James in 1952, Nobel Prize 1952) gave chemists their first routine tool for separating and quantifying volatile mixtures that had previously required days of fractional distillation. The first commercial GC-MS debuted in 1965; IR and UV-Vis spectrophotometers became standard bench instruments in most chemistry departments by the mid-1950s (Perkin-Elmer had been manufacturing IR instruments since 1940). Nuclear magnetic resonance spectrometers entered chemistry departments starting in the late 1950s and became the primary structure-determination tool for organic chemists through the 1960s. These instruments accelerated structure determination from weeks of classical degradation chemistry to hours of spectral interpretation, releasing chemists from the most time-consuming routine identification work and allowing smaller teams to cover more chemical territory. The net effect on employment was expansionary: faster structure confirmation enabled faster synthesis programs, and faster programs meant more chemists could work productively in parallel.

  Effect on the work

  Each major instrument (GC, NMR, IR, MS) displaced specific manual wet-chemistry workflows but opened new research capability that absorbed the freed time. The 1960s-1970s pharmaceutical industry used the new speed of spectroscopic structure confirmation to accelerate drug discovery campaigns dramatically, growing the total chemist workforce throughout the period rather than shrinking it.

• HPLC autosamplers1980–1995

  HPLC autosamplers, computerized data systems, and laboratory robotics

  The 1980s brought computational data-reduction to the analytical chemistry bench. HPLC (high-performance liquid chromatography) autosamplers allowed a single analytical chemist to queue up many samples overnight that previously would have required manual injections over multiple days. Computerized data acquisition systems -- replacing chart recorders with workstations running chromatography software -- allowed results to be compared, integrated, and archived electronically. Laboratory robotics systems from Zymark could perform the entire sample preparation and NMR tube insertion sequence automatically. Zymark's 1980s systems reduced hands-on bench time for routine analytical workflows for the specific tasks they automated. The net effect was a radical reorganization of the analytical chemistry department: fewer technician hours per batch, faster turnaround for synthesis programs, and a shift in the analytical chemist's role from hands-on sample handling toward interpretation, method development, and data oversight.

  Effect on the work

  Laboratory automation in the 1980s did reduce the number of chemical technicians and junior analytical chemists needed per sample batch, particularly in pharmaceutical QC. It did not reduce the total demand for bench chemists -- it redirected demand toward method development, instrument qualification, and data interpretation, roles requiring more chemical expertise, not less.

• Molecular modeling1990–2015

  Molecular modeling and computational chemistry: Gaussian, HyperChem, DFT, and virtual screening

  Gaussian (originating from John Pople's group at Carnegie Mellon and released commercially from 1970, with PC-accessible versions from the late 1980s) and HyperChem (1992) brought quantum mechanical structure and energy calculations to the individual chemist's desktop. Through the 1990s, density functional theory (DFT) became practical on workstations fast enough to give useful energy and geometry predictions for drug-sized molecules. Medicinal and materials chemists could now generate and evaluate virtual molecular structures before committing to synthesis. High-throughput virtual screening -- evaluating millions of candidate molecules by predicted binding affinity before purchasing or synthesizing any of them -- emerged as a standard early-stage tool in pharmaceutical discovery by the mid-2000s. The Schrödinger Glide docking suite (commercial from 2001) became the industry standard for structure-based virtual screening. This shifted the early phase of a drug discovery campaign from physical assay of compound libraries toward computational pre-filtering, dramatically increasing the hit rate of the molecules that reached the bench.

  Effect on the work

  Virtual screening did not eliminate the medicinal chemist; it shifted the bottleneck from "synthesize and test everything" to "synthesize and test the AI-predicted winners." Chemistry departments grew a new specialty -- computational chemist, cheminformatics scientist -- without contracting the synthetic side, because faster design cycles required more synthesis throughput, not less.

• Electronic lab notebooks2015–2022

  Electronic lab notebooks (ELN), cloud data platforms, and high-throughput screening automation

  The mid-2010s brought the digitization of the chemistry lab itself. Electronic lab notebooks -- most prominently Benchling (founded 2012, reaching 70% of top-20 pharma by 2025) -- replaced paper notebooks with searchable, auditable, structured data records. This was not merely administrative convenience: structured ELN data made synthesis records machine-queryable for the first time, enabling a chemistry department to search its entire experimental history for precedents, reagent usage patterns, and yield statistics that previously lived in unindexed paper notebooks on shelves. Simultaneously, acoustic dispensing systems (Echo liquid handlers), automated synthesis platforms, and plate-reader arrays brought higher throughput to medicinal and materials chemistry workflows. The volume of compounds a team could characterize per week expanded dramatically without a proportional increase in headcount. The regulatory significance was also important: 21 CFR Part 11 compliant ELN records dramatically simplified FDA inspection preparation for pharmaceutical laboratories.

• AI-native chemistry2022–present

  AI-native chemistry: IBM RXN retrosynthesis, AlphaFold 3, Schrodinger FEP+, Citrine active-learning DOE

  The 2022-2026 wave of AI tools represents the most concentrated transformation of the chemist's workflow since the introduction of NMR. IBM RXN for Chemistry (trained on millions of reactions, commercially available since 2018 and widely adopted in pharma R&D by 2022) generates multi-step retrosynthetic routes from a target molecule in minutes -- work that previously occupied a medicinal chemist for a week of SciFinder searches. AlphaFold 3 (DeepMind, May 2024) predicts protein-ligand-DNA co-complex structures with atomic-level accuracy, giving structural biologists and medicinal chemists structural hypotheses that previously required months of X-ray crystallography campaigns. Schrödinger FEP+ free energy perturbation provides quantitative binding and solubility predictions that compress hit-to-lead timelines significantly. Citrine Informatics active-learning DOE reduces the number of experiments required to navigate multi-dimensional materials composition spaces by 3-10x. Mestrelab Mnova AI automates NMR peak assignment and structure verification for routine compound classes, reducing spectral processing time by 60-80%. Taken together, the AI-native chemist -- fluent in both bench technique and this toolstack -- completes synthesis campaigns faster than the pre-AI baseline, according to early ACS survey data. The role is augmenting at extraordinary speed, and demand is growing: BLS projects +5% through 2034, driven by pharma R&D expansion (ADCs, PROTACs, RNA therapeutics), battery materials, and semiconductor chemicals.

  Effect on the work

  No net employment contraction is projected. The AI tools are compressing cycle time within existing research programs, not eliminating the programs.